University of Southern California Dissertations and Theses

Indoor air quality for human health in residential buildings

(USC Thesis Other)

Indoor air quality for human health in residential buildings

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content

Indoor Air Quality:
Indoor Air Quality for Human Health in Residential Buildings

By

Minghuan Gong

A Thesis Presented to the
FACULTY OF THE USC SCHOOL OF ARCHITECTURE
UNIVERSITY OF SOUTHERN CALIFORNIA
In Partial Fulfillment of the
Requirements for the Degree
MASTER OF BUILDING SCIENCE

August 2022

Copyright 2020 Minghuan Gong

i

ACKNOWLEDGEMENTS

I would like to thank my parents for supporting me throughout the thesis research all the time
and my friends for encouraging me, and I also would like to thank all the participants for
providing their valuable data for this research.

I would really appreciate professor Joon-Ho Choi (joonhoch@usc.edu USC School of
Architecture), Rima Habre (habre@usc.edu Keck School of Medicine), Kyle Konis
(kkonis@usc.edu USC School of Architecture), and Dong Yoon Park (dypark@kaist.ac.kr
USC School of Architecture) for providing valuable suggestions, patient instructions and
encouragements.

I would also like to thank Yixiao Li for providing technical support in the machine learning
part of data analysis.

ii

TABLE OF CONTENTS
ACKNOWLEDGEMENTS ................................................................................................... i
LIST OF TABLES ................................................................................................................ v
LIST OF FIGURES ........................................................................................................... viii
ABSTRACT ........................................................................................................................ xi
Chapter 1: INTRODUCTION ............................................................................................... 1
1.1 Indoor Environmental Quality ......................................................................................... 1
1.1.1 Importance of indoor environment ......................................................................... 1
1.1.2 Importance of indoor air quality ............................................................................. 1
1.1.3 Importance of indoor ventilation ............................................................................ 2
1.1.4 Importance of thermal comfort............................................................................... 2
1.2 Indoor Environmental Issues ........................................................................................... 3
1.2.1 Indoor environment under COVID-19 ................................................................... 3
1.2.2 Inadequate current indoor air quality standards ...................................................... 3
1.2.3 Impacts of indoor air pollution on individual occupants ......................................... 4
1.3 Biometric signals as indicators of human physiological states ......................................... 5
1.3.1 Skin Temperature .................................................................................................. 5
1.3.2 Heart Rate ............................................................................................................. 5
1.3.3 Electroencephalogram (EEG) ................................................................................ 6
1.3.4 Electrical activity of skin (EDA) ............................................................................ 6
1.4 Machine Learning techniques .......................................................................................... 7
1.4.1 Linear Regression .................................................................................................. 8
1.4.2 Random Forest ...................................................................................................... 8
1.4 Summary ......................................................................................................................... 8
Chapter 2: LITERATURE REVIEW ..................................................................................... 9
2.1 Impact of indoor air pollution on human being ................................................................ 9
2.1.1 Importance of maintaining good indoor air quality ................................................. 9
2.1.2 Impacts of indoor air quality on work efficiency .................................................. 10
2.1.3 Important factors of indoor air quality under the Covid-19 ................................... 11
2.1.3.1 PM 2.5 as an important indicator ................................................................ 11
2.1.3.2 CO2 as an important factor ......................................................................... 12
2.2 Impact of indoor air pollution on individual occupants .......................................... 12
2.3 Machine learning in building science domain ........................................................ 12
2.4 Indoor environmental quality and human physiological responses ......................... 13
2.4.1 Human physiological responses and bio-signals in building science domain ........ 13
2.4.2 The correlation between indoor air quality and human physiological responses ... 13
2.5 Effective control/Energy efficient control ...................................................................... 15
2.5.1 Improve indoor air quality/overcome the air pollution.......................................... 15
2.5.3 Energy saving ...................................................................................................... 16
2.6 Summary ....................................................................................................................... 17

iii

Chapter 3: METHODOLOGY ............................................................................................ 17
3.1 Project Design ....................................................................................................... 17
3.2 Hardware and Software ......................................................................................... 19
3.2.1 Sensors for IEQ measurements ............................................................................ 20
3.2.1.1 Sensor for collecting air temperature and CO2 concentration ..................... 20
3.2.1.2 Sensor for collecting Particulate Matters concentration .............................. 21
3.2.1.3 Sensor for collecting lighting ..................................................................... 22
3.2.2.4 Sensor for collecting acoustic ..................................................................... 22
3.2.2 Sensors for collecting human bio-signals ............................................................. 23
3.2.2.1 Sensor for collecting heart rate and stress level ........................................... 23
3.2.2.2 Sensor for collecting skin temperature and EDA ........................................ 24
3.2.3 Survey ................................................................................................................. 24
3.3 Data collection ...................................................................................................... 27
3.3.1 Participants’ personal information data ................................................................ 27
3.3.2 Participants’ physiological responses data ............................................................ 27
3.3.3 Environmental parameters data ............................................................................ 28
3.3.4 Survey data .......................................................................................................... 29
3.4 Machine learning ................................................................................................... 29
3.4.1 Data Organization ................................................................................................ 29
3.4.2 Data Analysis ...................................................................................................... 29
3.5 Summary ............................................................................................................... 30
Chapter 4: DATA ANALYSIS............................................................................................ 31
4.1 Pearson Correlation Coefficient between all parameters......................................... 31
4.4.1 Participant 1 ........................................................................................................ 32
4.4.2 Participant 2 ........................................................................................................ 34
4.4.3 Participant 3 ........................................................................................................ 37
4.2 Investigate the most relevant bio-signal to IAQ parameters .................................... 39
4.2.1 Participant 1 ........................................................................................................ 39
4.2.2 Participant 2 ........................................................................................................ 40
4.2.3 Participant 3 ........................................................................................................ 41
4.3 Investigate the most relevant bio-signal to IAQ Evaluation............................................ 43
4.3.1 Participant 1 ........................................................................................................ 43
4.3.2 Participant 2 ........................................................................................................ 46
4.3.3 Participant 3 ........................................................................................................ 50
4.4 Investigate the most relevant factors (including bio-signals, indoor temp, indoor
RH and IAQ parameters) to IAQ Evaluation ....................................................................... 53
4.4.1 Participant 1 ........................................................................................................ 53
4.4.2 Participant 2 ........................................................................................................ 56
4.4.3 Participant 3 ........................................................................................................ 60
4.5 Investigate the most relevant factors (including bio-signals, indoor temp and indoor
RH) to thermal comfort ....................................................................................................... 64
4.5.1 Participant 1 ........................................................................................................ 64
4.5.2 Participant 2 ........................................................................................................ 68

iv

Chapter 5: RESULTS.......................................................................................................... 73
5.1 Physiological signal most relevant to IAQ parameters ........................................... 73
5.1.1Results based on Pearson Correlation analysis ...................................................... 73
5.1.1.1 The most correlated bio-signal to indoor CO2 concentration ....................... 73
5.1.1.2The most correlated bio-signal to indoor PM2.5 concentration .................... 76
5.1.2 Results based on Stepwise Linear Regression ...................................................... 79
5.1.2.1 The most significant bio-signal to indoor CO2 concentration ..................... 79
5.1.2.2 The most significant bio-signal to indoor PM2.5 concentration .................. 83
5.2 More relevant physiological signals to IAQ Evaluation ......................................... 86
5.2.1 Results based on Stepwise Linear Regression ...................................................... 86
5.2.2 Results based on Random Forest result ................................................................ 90
5.3 More relevant factors to IAQ Evaluation ............................................................... 91
5.3.1 Results based on Stepwise linear regression ......................................................... 91
5.3.2 Results based on Random Forest result ................................................................ 93
5.4 Some specific relationship about two bio-signals (HR and stress level) and IAQ ... 95
5.4.1 The relationship between two bio-signals (HR and stress level) and IAQ parameters
95
5.4.1.1 The relationship between heart rate and IAQ parameters ............................ 95
5.4.1.2 The relationship between stress level and IAQ parameters.......................... 96
5.4.2 The relationship between two bio-signals (HR and stress level) and IAQ Evaluation
97
5.4.2.1 The relationship between heart rate and IAQ Evaluation ................................... 97
5.5 Most relevant factors to thermal comfort ............................................................... 97
5.5.1 Results based on Stepwise linear regression ......................................................... 97
5.5.2 Results based on random forest ............................................................................ 99
Chapter 6: DISCUSSION, LIMITATION AND FUTURE WORK ................................... 101
6.1 Conclusion and discussion ................................................................................... 101
6.2 Limitation and future work .................................................................................. 102
6.2.1 short term problems ........................................................................................... 102
6.2.2 long term problems ............................................................................................ 103
6.2.3 Additional future work ....................................................................................... 103
REFERENCES ................................................................................................................. 104

v

LIST OF TABLES
Table 3. 1 IEQ guidelines ................................................................................................ 20
Table 3. 2 HOBO MX1102 Details .................................................................................. 21
Table 3. 3 Pa- II- SD Details ............................................................................................ 22
Table 3. 4 Empatica Embrace 2 Details ............................................................................ 22
Table 3. 5 Empatica Embrace 2 Details ............................................................................ 23
Table 3. 6 Garmin Vivosmart 3 Details ............................................................................ 23
Table 3. 7 Empatica Embrace 2 Details ............................................................................ 24
Table 4. 1 Descriptive Statistics Table for Participant 1 ................................................... 32
Table 4. 2 Descriptive Statistics Table for Participant 2 ................................................... 35
Table 4. 3 Descriptive Statistics Table for Participant 3 ................................................... 37
Table 4. 4 Summary of relevant bio-signals based on two different methods - Participant
1 ............................................................................................................................... 46
Table 4. 5 Summary of accuracy of different IAQ_Evaluation prediction model -
Participant 1 .............................................................................................................. 46
Table 4. 6 Summary of relevant bio-signals based on two different methods - Participant
2 ............................................................................................................................... 49
Table 4. 7 Summary of accuracy of different IAQ_Evaluation prediction model -
Participant 2 .............................................................................................................. 49
Table 4. 8 Summary of relevant bio-signals based on two different methods - Participant
3 ............................................................................................................................... 52
Table 4. 9 Summary of accuracy of different IAQ_Evaluation prediction model -
Participant 3 .............................................................................................................. 52
Table 4.10 Summary of relevant parameters based on two different methods - Participant
1 ............................................................................................................................... 56
Table 4.11 Summary of accuracy of different IAQ_Evaluation prediction model -
Participant 1 .............................................................................................................. 56
Table 4.12 Summary of relevant parameters based on two different methods - Participant
2 ............................................................................................................................... 59
Table 4.13 Summary of relevant parameters based on two different methods - Participant
3 ............................................................................................................................... 63
Table 4. 14 Summary of accuracy of different IAQ_Evaluation prediction model -
Participant 3 .............................................................................................................. 64
Table 4. 15 Summary of relevant parameters to thermal comfort - Participant 1 ............... 67
Table 4. 16 Summary of accuracy of different thermal comfort prediction model -
Participant 1 .............................................................................................................. 68
Table 4. 17 Summary of relevant parameters based on two different methods -
Participant 2 .............................................................................................................. 71
Table 4.18 Summary of accuracy of different thermal comfort prediction model -
Participant 2 .............................................................................................................. 72
Table 5. 1Rank of correlation between bio-signals and CO2 for each participant .............. 73
Table 5. 2 Number of total times each bio-signal appears in the first rank (CO2) .............. 74

vi

Table 5. 3 Score distribution for each bio-signal .............................................................. 74
Table 5. 4 Number of total times each bio-signal appears in the first rank in male group .. 75
Table 5. 5 Number of total times each bio-signal appears in the first rank in male group .. 75
Table 5. 6 Rank of correlation between bio-signals and PM2.5 for each participant ......... 76
Table 5. 7 Number of total times each bio-signal appears in the first rank (PM2.5) .......... 77
Table 5. 8 Score distribution for each bio-signal .............................................................. 77
Table 5. 9 Number of total times each bio-signal appears in the first rank in male group
(PM2.5) .................................................................................................................... 78
Table 5. 10 Number of total times each bio-signal appears in the first rank in female
group ........................................................................................................................ 78
Table 5. 11 Rank of significance of four bio-signals to indoor CO2 concentration ............ 80
Table 5. 12 Number of total times each bio-signal appears in the first rank (CO2) ............ 80
Table 5. 13 Score distribution for each bio-signal (CO2) .................................................. 80
Table 5.14 Number of total times each bio-signal appears in the first rank in male group
(CO2) ........................................................................................................................ 81
Table 5. 15 Number of total times each bio-signal appears in the first rank in female
group (CO2) .............................................................................................................. 82
Table 5. 16 Rank of significance of four bio-signals to indoor PM2.5 concentration ........ 83
Table 5. 17 Number of total times each bio-signal appears in the first rank (PM2.5) ........ 83
Table 5. 18 Score distribution for each bio-signal ............................................................ 83
Table 5. 19 Number of total times each bio-signal appears in the first rank in male group
(PM2.5) .................................................................................................................... 84
Table 5. 20 Number of total times each bio-signal appears in the first rank in female
group (PM2.5)........................................................................................................... 85
Table 5. 21 Rank of significance of important bio-signals of each participant to IEQ
Evaluation - Stepwise ................................................................................................ 87
Table 5. 22 Number of total times each bio-signal appears in the first rank (IAQ
Evaluation) ............................................................................................................... 87
Table 5. 23 Score distribution for each bio-signal (IEQ Evaluation) ................................. 87
Table 5. 24 Number of total times each bio-signal appears in the first rank in male group
(IAQ Evaluation) ...................................................................................................... 88
Table 5. 25 Number of total times each bio-signal ranks first in female group (IAQ
Evaluation) ............................................................................................................... 89
Table 5. 26 Rank of significance of important bio-signals of each participant to IEQ
Evaluation – Random Forest ..................................................................................... 90
Table 5. 27 Number of total times each bio-signal appears in the first rank (IAQ
Evaluation) ............................................................................................................... 90
Table 5. 28 Score distribution for each bio-signal (IEQ Evaluation) ................................. 91
Table 5. 29 Rank of significance of significant parameters to IEQ Evaluation and the
prediction model accuracy of each subject – Stepwise linear regression .................... 92
Table 5.30 Number of total times each eight parameters appears in the first rank (IAQ
Evaluation) ............................................................................................................... 92
Table 5. 31 Score distribution for each eight parameters (IEQ Evaluation) ....................... 92

vii

Table 5. 32 Rank of significance of significant parameters to IEQ Evaluation and the
model accuracy of each subject – Random Forest ...................................................... 94
Table 5.33 Number of total times each eight parameters appears in the first rank (IAQ
Evaluation) ............................................................................................................... 94
Table 5. 34 Pearson correlation coefficient between HR and IAQ parameters for each
participant ................................................................................................................. 96
Table 5. 35 Number of positive and negative correlations between HR and IAQ
parameters................................................................................................................. 96
Table 5.36 Pearson correlation coefficient between stress level and IAQ parameters for
each participant ......................................................................................................... 96
Table 5. 37 Number of positive and negative correlations between stress level and IAQ
parameters................................................................................................................. 96
Table 5. 38 Pearson correlation coefficient between two bio-signals (HR and stress level)
and IEQ Evaluation for each participant .................................................................... 97
Table 5.39 Rank of significance of significant parameters of each subject to thermal
comfort - Stepwise .................................................................................................... 98
Table 5. 40 Number of total times each eight parameters appears in the first rank (IAQ
Evaluation) ............................................................................................................... 98
Table 5. 41 Score distribution for each eight parameters (thermal comfort) ...................... 98
Table 5. 42 Rank of significance of significant parameters to thermal comfort and the
model accuracy of each subject – Random Forest ...................................................... 99

viii

LIST OF FIGURES
Figure 1. 1 Illustration of SCR, SCL and EDA Peaks. (Retrieved from
https://imotions.com/blog/skin-conductance-response/) .............................................. 7
Figure 2. 1 Burden of disease from poor IAQ by pollutant - EU26 - 2010. (Retrieved
from https://www.e3s-
conferences.org/articles/e3sconf/abs/2018/24/e3sconf_solina2018_00133/e3sconf_s
olina2018_00133.html) .............................................................................................. 9
Figure 2. 2 Total burden of disease - EU26 - 2010. (Retrieved from https://www.e3s-
conferences.org/articles/e3sconf/abs/2018/24/e3sconf_solina2018_00133/e3sconf_s
olina2018_00133.html) ............................................................................................. 10
Figure 2. 3 The floor plan of the typical type of sample homes showing the location of
humidity, temperature and CO2 data logger (Lutron data loggers, MCH-383SD) and
binary status logger (UX90-001, Onset Computer) (Retrieved from https://journals-
sagepub-com.libproxy2.usc.edu/doi/full/10.1177/1420326X20927070 ). .................. 16
Figure 3. 1Whole progress ............................................................................................... 18
Figure 3. 2 Methodology diagram .................................................................................... 19
Figure 3. 3 HOBO MX1102 ............................................................................................. 20
Figure 3. 4 Pa- II- SD....................................................................................................... 21
Figure 3. 5 Dr.meter LX1330B Digital Illuminance Light Meter ...................................... 22
Figure 3. 6 PCE-SDL 1 .................................................................................................... 23
Figure 3. 7 Garmin Vivosmart 3....................................................................................... 23
Figure 3. 8 Empatica Embrace 2 ...................................................................................... 24
Figure 3. 9 Indoor air quality Evaluation Survey .............................................................. 25
Figure 3. 10 Indoor thermal comfort survey ..................................................................... 26
Figure 3. 11 All data collected from the first
participant…………………………………………………29
Figure. 4 1 Heat Map of Pearson Correlation between IAQ Parameters and Bio-signals -
Participant 1 .............................................................................................................. 33
Figure. 4 2 Scatter plots of Linear regression between IAQ Parameters and Bio-signals -
Participant 1 .............................................................................................................. 34
Figure. 4 3 Heat Map of Pearson Correlation between other IEQ Parameters and survey
data and Bio-signals -Participant 1 ............................................................................ 34
Figure. 4 4 Heat Map of Correlation between IAQ Parameters and Bio-signals -
Participant 2 .............................................................................................................. 35
Figure. 4 5 Scatter plots of Linear regression between IAQ Parameters and Bio-signals -
Participant 2 .............................................................................................................. 36
Figure. 4 6 Heat Map of Correlation between other IEQ Parameters and survey data and
Bio-signals -Participant 2 .......................................................................................... 36
Figure. 4 7 Heat Map of Correlation between IAQ Parameters and Bio-signals -
Participant 3 .............................................................................................................. 38
Figure. 4 8 Scatter plots of Linear regression between IAQ Parameters and Bio-signals -
Participant 3 .............................................................................................................. 38

ix

Figure. 4 9 Heat Map of Correlation between other IEQ Parameters and Bio-signals -
Participant 3 .............................................................................................................. 39
Figure. 4 10 Stepwise Linear Regression Analysis Result (Dependent variable: CO 2,
Independent variables: four bio-signals) -Participant 1 .............................................. 40
Figure. 4 11 Stepwise Linear Regression Analysis Result (Dependent variable: PM2.5,
Independent variables: four bio-signals) -Participant 1 .............................................. 40
Figure. 4 12 Stepwise Linear Regression Analysis Result (Dependent variable: CO 2,
Independent variables: four bio-signals) -Participant 1 .............................................. 41
Figure. 4 13 Stepwise Linear Regression Analysis Result (Dependent variable: PM2.5,
Independent variables: four bio-signals) -Participant 2 .............................................. 41
Figure. 4 14 Stepwise Linear Regression Analysis Result (Dependent variable: CO2,
Independent variables: four bio-signals) -Participant 3 .............................................. 42
Figure. 4 15 Stepwise Linear Regression Analysis Result (Dependent variable: PM2.5,
Independent variables: four bio-signals) -Participant 3 .............................................. 42
Figure. 4 16 Stepwise Linear Regression Analysis Result (Dependent variable: IAQ
Evaluation, Independent variables: heart rate, stress level, skin temperature, EDA) -
Participant 1 .............................................................................................................. 43
Figure. 4 17 Random Forest Algorithm and Accuracy of the model (Use stress level and
skin temperature to predict IAQ Evaluation, use cross validation to test accuracy)-
Participant 1 .............................................................................................................. 43
Figure. 4 18 Random Forest Algorithm and Accuracy of the model (Use all bio-signals to
predict IAQ Evaluation, use cross validation to test accuracy)-Participant 1 .............. 44
Figure. 4 19 Random Forest Algorithm for using first day to train a model and last day
data to test accuracy of the model (Use stress level and skin temperature to predict
IAQ Evaluation)-Participant 1 ................................................................................... 44
Figure. 4 20 Random Forest Algorithm for using first day to train a model and last day
data to test accuracy of the model (Use all bio-signals to predict IAQ Evaluation)-
Participant 1 .............................................................................................................. 44
Figure. 4 21 Feature importance (four bio-signals) graph from Random Forest -
Participant 1 .............................................................................................................. 45
Figure. 4 22 Random Forest Algorithm and Accuracy of the model (Use heart rate and
stress level to predict IAQ Evaluation, use cross validation to test accuracy)-
Participant 1 .............................................................................................................. 45
Figure. 4 23 Random Forest Algorithm for using first day to train a model and last day
data to test accuracy of the model (Use stress level and skin temperature to predict
IAQ Evaluation)-Participant 1 ................................................................................... 46
Figure. 4 24 Stepwise Linear Regression Analysis Result (Dependent variable: IAQ
Evaluation, Independent variables: heart rate, EDA, skin temperature, and stress
level) -Participant 2 ................................................................................................... 47
Figure. 4 25 Random Forest Algorithm and Accuracy of the model (Use heart rate to
predict IAQ Evaluation)-Participant 2 ....................................................................... 47
Figure. 4 26 Random Forest Algorithm and Accuracy of the model (Use all bio-signals to
predict IAQ Evaluation)-Participant 2 ....................................................................... 47

x

Figure. 4 27 Random Forest Algorithm for using first day to train a model and last day
data to test accuracy of the model (Use heart rate to predict IAQ Evaluation)-
Participant 2 .............................................................................................................. 48
Figure. 4 28 Random Forest Algorithm for using first day to train a model and last day
data to test accuracy of the model (Use all bio-signals to predict IAQ Evaluation)-
Participant 2 .............................................................................................................. 48
Figure. 4 29 Feature importance (four bio-signals) graph from Random Forest -
Participant 2 .............................................................................................................. 48
Figure. 4 30 Random Forest Algorithm and Accuracy of the model (Use EDA to predict
IAQ Evaluation, use cross validation to test accuracy)-Participant 2 .......................... 49
Figure. 4 31 Random Forest Algorithm for using first day to train a model and last day
data to test accuracy of the model (Use EDA to predict IAQ Evaluation)-Participant
2 ............................................................................................................................... 49
Figure. 4 32 Stepwise Linear Regression Analysis Result (Dependent variable: IAQ
Evaluation, Independent variables: heart rate, stress level, skin temperature, EDA) -
Participant 3 .............................................................................................................. 50
Figure. 4 33 Random Forest Algorithm and Accuracy of the model (Use all bio-signals to
predict IAQ Evaluation)-Participant 3 ....................................................................... 50
Figure. 4 34 Random Forest Algorithm for using first day to train a model and last day
data to test accuracy of the model (Use heart rate to predict IAQ Evaluation)-
Participant 2 .............................................................................................................. 51
Figure. 4 35 Feature importance (four bio-signals) graph from Random Forest -
Participant 3 .............................................................................................................. 51
Figure. 4 36 Random Forest Algorithm and Accuracy of the model (Use EDA and stress
level to predict IAQ Evaluation, use cross validation to test accuracy)-Participant 3 .. 51
Figure. 4 37 Random Forest Algorithm for using first day to train a model and last day
data to test accuracy of the model (Use EDA to predict IAQ Evaluation)-Participant
3 ............................................................................................................................... 52
Figure. 4 38 Stepwise Linear Regression Analysis Result (Dependent variable: IAQ
Evaluation, Independent variables: four bio-signals, indoor temperature, indoor
relative humidity, indoor CO2, and indoor PM 2.5) -Participant 1 .............................. 53
Figure. 4 39 Random Forest Algorithm and Accuracy of the model (Use indoor PM2.5,
stress level, indoor CO2, and indoor temperature to predict IAQ Evaluation)-
Participant 1 .............................................................................................................. 53
Figure. 4 40 Random Forest Algorithm and Accuracy of the model (Use all eight
parameters to predict IAQ Evaluation)-Participant 1 ................................................. 54
Figure. 4 41 Random Forest Algorithm for using first day to train a model and last day
data to test accuracy of the model (Use PM2.5, stress level, CO2, and indoor
temperature to predict IAQ Evaluation)-Participant 1 ................................................ 54
Figure. 4 42 Random Forest Algorithm for using first day to train a model and last day
data to test accuracy of the model (Use all eight parameters to predict IAQ
Evaluation)-Participant 1 ........................................................................................... 54
Figure. 4 43 Feature importance (eight parameters) graph from Random Forest -

xi

Participant 1 .............................................................................................................. 55
Figure. 4 44 Random Forest Algorithm and Accuracy of the model (Use PM2.5, stress
level, CO2 and heart rate to predict IAQ Evaluation, use cross validation to test
accuracy)-Participant 1 .............................................................................................. 55
Figure. 4 45 Random Forest Algorithm for using first day to train a model and last day
data to test accuracy of the model (Use PM2.5, stress level, CO 2 and heart rate to
predict IAQ Evaluation)-Participant 1 ....................................................................... 56
Figure. 4 46 Stepwise Linear Regression Analysis Result (Dependent variable: IAQ
Evaluation, Independent variables: four bio-signals, temperature, relative humidity,
CO2, and PM 2.5) -Participant 2 ................................................................................ 57
Figure. 4 47 Random Forest Algorithm and Accuracy of the model (Use indoor
temperature, indoor PM2.5, indoor relative humidity, and EDA to predict IAQ
Evaluation)-Participant 2 ........................................................................................... 57
Figure. 4 48 Random Forest Algorithm and Accuracy of the model (Use all eight
parameters to predict IAQ Evaluation)-Participant 2 ................................................. 57
Figure. 4 49 Random Forest Algorithm for using first day to train a model and last day
data to test accuracy of the model (Use PM2.5, stress level, CO2, and indoor
temperature to predict IAQ Evaluation)-Participant 2 ................................................ 58
Figure. 4 50 Random Forest Algorithm for using first day to train a model and last day
data to test accuracy of the model (Use all eight parameters to predict IAQ
Evaluation)-Participant 2 ........................................................................................... 58
Figure. 4 51 Feature importance (eight parameters) graph from Random Forest -
Participant 2 .............................................................................................................. 58
Figure. 4 52 Random Forest Algorithm and Accuracy of the model (Use EDA, CO2,
indoor temperature, and PM2.5 to predict IAQ Evaluation, use cross validation to
test accuracy)-Participant 2 ....................................................................................... 59
Figure. 4 53 Random Forest Algorithm for using first day to train a model and last day
data to test accuracy of the model (Use EDA, CO2, indoor temperature, and PM2.5
to predict IAQ Evaluation)-Participant 2 ................................................................... 59
Figure. 4 54 Stepwise Linear Regression Analysis Result (Dependent variable: IAQ
Evaluation, Independent variables: four bio-signals, temperature, relative humidity,
CO2, and PM 2.5) -Participant 3 ............................................................................... 60
Figure. 4 55 Random Forest Algorithm and Accuracy of the model (Use indoor relative
humidity, stress level, skin temperature, and indoor temperature to predict IAQ
Evaluation)-Participant 3 ........................................................................................... 61
Figure. 4 56 Random Forest Algorithm and Accuracy of the model (Use all eight
parameters to predict IAQ Evaluation)-Participant 3 ................................................. 61
Figure. 4 57 Random Forest Algorithm for using first day to train a model and last day
data to test accuracy of the model (Use indoor relative humidity, stress level, skin
temperature, and indoor temperature to predict IAQ Evaluation) - Participant 3 ........ 61
Figure. 4 58 Random Forest Algorithm for using first day to train a model and last day
data to test accuracy of the model (Use all eight parameters to predict IAQ
Evaluation)-Participant 3 ........................................................................................... 62

xii

Figure. 4 59 Feature importance (eight parameters) graph from Random Forest -
Participant 3 .............................................................................................................. 62
Figure. 4 60 Random Forest Algorithm and Accuracy of the model (Use indoor relative
humidity, PM2.5, CO2, and skin temperature to predict IAQ Evaluation, use cross
validation to test accuracy)-Participant 3 ................................................................... 63
Figure. 4 61 Random Forest Algorithm for using first day to train a model and last day
data to test accuracy of the model (Use EDA, CO2, indoor temperature, and PM2.5
to predict IAQ Evaluation)-Participant 3 ................................................................... 63
Figure. 4 62 Stepwise Linear Regression Analysis Result (Dependent variable: Thermal
Comfort, Independent variables: four bio-signals, temperature, relative humidity) -
Participant 1 .............................................................................................................. 64
Figure. 4 63 Random Forest Algorithm and Accuracy of the model (Use indoor
temperature, stress level, EDA, and skin temperature to predict thermal comfort, use
cross validation test accuracy)-Participant 1 .............................................................. 65
Figure. 4 64 Random Forest Algorithm and Accuracy of the model (Use all six
parameters to predict thermal comfort, use cross validation test accuracy)-Participant
1 ............................................................................................................................... 65
Figure. 4 65 Random Forest Algorithm for using first day to train a model and last day
data to test accuracy of the model (Use all six parameters to predict thermal
comfort)-Participant 1 ............................................................................................... 66
Figure. 4 66 Feature importance (six parameters) graph from Random Forest - Participant
1 ............................................................................................................................... 66
Figure. 4 67 Random Forest Algorithm and Accuracy of the model (Use indoor
temperature, skin temperature, indoor relative humidity and EDA to predict thermal
comfort, use cross validation to test accuracy)-Participant 1 ...................................... 67
Figure. 4 68 Random Forest Algorithm for using first day to train a model and last day
data to test accuracy of the model (Use indoor temperature, skin temperature, indoor
relative humidity and EDA to predict thermal comfort)-Participant 1 ........................ 67
Figure. 4 69 Stepwise Linear Regression Analysis Result (Dependent variable: Thermal
Comfort, Independent variables: four bio-signals, temperature, relative humidity) -
Participant 2 .............................................................................................................. 68
Figure. 4 70 Random Forest Algorithm and Accuracy of the model (Use skin
temperature, indoor temperature, EDA, and stress level to predict thermal comfort)-
Participant 2 .............................................................................................................. 69
Figure. 4 71 Random Forest Algorithm and Accuracy of the model (Use all six
parameters to predict thermal comfort)-Participant 2 ................................................. 69
Figure. 4 72 Random Forest Algorithm for using first day to train a model and last day
data to test accuracy of the model (Use skin temperature, indoor temperature, EDA,
and stress level to predict thermal comfort) - Participant 2 ........................................ 69
Figure. 4 73 Random Forest Algorithm for using first day to train a model and last day
data to test accuracy of the model (Use all six parameters to predict thermal
comfort)-Participant 2 ............................................................................................... 70
Figure. 4 74 Feature importance (six parameters) graph from Random Forest - Participant

xiii

2 ............................................................................................................................... 70
Figure. 4 75 Random Forest Algorithm and Accuracy of the model (Use indoor
temperature, indoor relative humidity, EDA, and skin temperature to predict thermal
comfort, use cross validation to test accuracy)-Participant 2 ...................................... 71
Figure. 4 76 Random Forest Algorithm for using first day to train a model and last day
data to test accuracy of the model (Use indoor temperature, indoor relative humidity,
EDA, and skin temperature to predict thermal comfort)-Participant 2 ........................ 71
Figure 5.1.1 Total scores for each bio-signal – CO2 ........................................................... 74
Figure 5.1.2 Total scores for each bio-signal in male group – CO2 .................................. 75
Figure 5.1.3 Total scores for each bio-signal in female group – CO2 ................................ 76
Figure 5.1.4 Comparison chart of total scores for each bio-signals in male and female
groups – CO2 ............................................................................................................ 76
Figure 5.1.5 Total score for each bio-signal – PM2.5 ....................................................... 77
Figure 5.1.6 Total scores for each bio-signal in male group – PM2.5 ............................... 78
Figure 5.1.7 Total scores for each bio-signal in female group – PM2.5 ............................ 79
Figure 5.1.8 Comparison chart of total scores for each bio-signals in male and female
groups – PM2.5 ......................................................................................................... 79
Figure 5.1.9 Total scores for each bio-signal – CO2 ......................................................... 81
Figure 5.1.10 Total scores for each bio-signal in male group – CO2 ................................. 81
Figure 5.1.11 Total scores for each bio-signal in female group – CO2 .............................. 82
Figure 5.1.12 Comparison chart of total scores for each bio-signals in male and female
groups – CO2 ............................................................................................................ 82
Figure 5.1.13 Total scores for each bio-signal – PM2.5 .................................................... 84
Figure 5.1.14 Total score for each bio-signal in male group – PM2.5 ............................... 84
Figure 5.1.15 Total scores for each bio-signal in female group – PM2.5........................... 85
Figure 5.1.16 Comparison chart of total scores for each bio-signals in male and female groups
– PM2.5
……………………………………………………………………………………………
………… ...86
Figure 5.2.1 Total scores for each bio-signal – IAQ Evaluation ........................................ 88
Figure 5.2.2 Total scores for each bio-signal in male group –IAQ Evaluation .................. 88
Figure 5.2.3 Total scores for each bio-signal in female group – IAQ Evaluation .............. 89
Figure 5.2.4 Comparison chart of total scores for each bio-signals in male and female
groups – IEQ Evaluation ........................................................................................... 89
Figure 5.2.5 Total scores for each bio-signal – IEQ Evaluation ........................................ 91
Figure 5.3.1 Total scores for each eight parameters – IAQ Evaluation ............................. 93
Figure 5.3.2 Total scores for each eight parameters – IAQ Evaluation ............................. 95
Figure 5.5.1 Total scores for each eight parameter – thermal comfort............................... 99
Figure 5.5.2 Total scores for each six parameter – thermal comfort ................................ 100

xi

ABSTRACT
Indoor environment quality (IEQ) is closely related to human life and health, and it has
significance for occupant wellness. Among IEQ, indoor air quality (IAQ) is increasingly
concerned and significantly affects our health in the built environment because everything we
breathe may affect our health. To determine whether IAQ really affects our health, well-
designed, scientifically valid studies are necessary. This experiment mainly explored the
correlation between indoor PM2.5 and CO2 on human physiological responses, investigating
the bio-signals most associated with indoor PM2.5 and CO2 and the bio- signals most related
to IAQ Evaluation and thermal comfort of all participants.
Twelve students from the University of Southern California volunteered to participate in
this study; the whole group's age range is between 25 ~35. The sensors measuring indoor and
outdoor climate data were installed indoors and outdoors for five days, and the human sensors
(two smartwatches) were used for collecting human bio-signals (heart rate, skin temperature,
stress level, and electrodermal activity). All the data was recorded every ten minutes while the
subjects were in the room. All participants needed to fill out IAQ Evaluation and thermal
comfort questionaries during the experiment every 2 hours. In analyzing the data of each
subject, Pearson Correlation, stepwise linear regression, and random forest were applied in data
analysis, prediction model creation, and model accuracy testing. Finally, common features
were found after comparing all the different individual analysis results. Among four bio-signals,
heart rate is most associated with indoor PM2.5 and CO 2 concentration, stress level is most
associated with IAQ Evaluation based on stepwise linear regression analysis, EDA is most
associated with IAQ Evaluation based on random forest analysis, skin temperature is most

xii

associated with thermal comfort. For other factors, relative humidity is most associated with
IAQ Evaluation based on stepwise linear regression analysis; indoor CO 2 concentration is most
associated with IAQ Evaluation based on random forest analysis.
Keywords: indoor environmental quality, indoor air quality, human physiological responses,
human health

xiii

Hypothesis:
The short-term changes in concentration of indoor CO2 and PM2.5 will have impacts on human
physiological responses (heart rate, skin temperature, stress level and electrodermal activity
(EDA).
Objectives
1. Identify the correlation between IAQ and human physiological responses at ten minutes
scale.
2. Investigate the more relevant bio-signal to the indoor air quality parameters (PM2.5 and
CO2) based on the analysis results.
3. Explore the more relevant bio-signals and other factors (including indoor temperature,
indoor humidity, IAQ parameters) to IAQ Evaluation and thermal comfort of each
participant and find the common features across all participants.

1

Chapter 1: INTRODUCTION
1.1 Indoor Environmental Quality
1.1.1 Importance of indoor environment
Indoor environmental quality (IEQ) refers to the quality of the built environment and is related
to the health and well-being of the people occupying the building space (Centers for Disease
Control and Prevention, n.d.). Also, IEQ is related to various factors that affect human life in
buildings, such as indoor air quality, acoustic, lighting, and damp conditions (Mujeebu, 2019);
these elements' relationship is complicated, which may cause long-term or short-term effects
on human health (Bluyssen, 2013). For example, too bright or dark indoor lighting will cause
visual discomfort. Also, some invisible air pollutants will go directly into the human's
bloodstream through inhalation, potentially leading to cardiovascular disease, asthma and even
cause cancer over the long term (Al horr et al., 2016)). Americans spend at least 90% of their
time indoors and indoor concentrations of some pollutants are usually two to five times higher
than typical outdoor concentrations (U.S. Environmental Protection Agency | US EPA, n.d.).
Therefore, building and maintaining a good indoor environment quality has great significance
for human health.
1.1.2 Importance of indoor air quality
Indoor air quality (IAQ) is regarded as the air quality inside and around buildings and structures,
and it is an essential factor to study the indoor comfort of occupants in buildings (U.S.
Environmental Protection Agency | US EPA, n.d.). Moreover, IAQ is essential to people's
health. Air pollution has significant impacts on most organs and systems of the human body
(Jiang et al., 2016). Numerous scientific evidence shows that air pollution can induce various
cardiovascular and respiratory diseases, such as chronic obstructive pulmonary disease and
asthma, and adversely affect the human nervous system, digestive system, and urinary system
(Marquès & Domingo, 2022). Also, air pollution can worsen respiratory diseases; some air
pollutants (Particulate matter, nitrogen oxides) could even increase the risk of lung cancer and
other lung diseases (Jiang et al., 2016). Therefore, the concentration of particulate matter can
affect IAQ and the health of occupants. In addition, Infants, the elderly, and people with
cardiovascular or respiratory diseases tend to spend more time indoors, and these people are
most vulnerable to the adverse effects of air pollution (U.S. Environmental Protection Agency
| US EPA, n.d.). Besides, according to World Health Organization, 20 percent of all-cause
deaths in Europe are due to environmental diseases (Agarwal et al., 2021), and air pollution

2

has been identified as the world's largest environmental cause of premature death (Forouzanfar
et al., 2016). Hence, people pay more and more attention to IAQ.
1.1.3 Importance of indoor ventilation
Ventilation of a residential building is critical to the building and its occupants as it provides
healthy indoor air, ensures energy efficiency, and maintains indoor air circulation, good IAQ,
and occupants' health and comfort. In general, IAQ contributes more to health and wellness
than outdoor air quality (Poirier et al., 2021), and the rate of transmission of the virus in the
indoor environment is much higher than that outside, possibly because of the longer exposure
time and the reduced level of indoor turbulence and dilution (Bhagat et al., 2020). In addition,
inadequate ventilation can increase indoor pollutant levels because not enough outdoor fresh
air is brought to dilute or remove emissions from indoor sources (U.S. Environmental
Protection Agency | US EPA, n.d.). However, proper ventilation will reduce indoor odor and
control CO2 and other gases; besides, the design and operation of ventilation systems
significantly affect IAQ (Bluyssen, 2013). Therefore, keeping the building ventilated and
ensuring good air quality is imperative.
1.1.4 Importance of thermal comfort
Based on the American Society of Heating, Refrigerating and Air Conditioning Engineers
(ASHRAE), thermal comfort is defined as the psychological state of expressing satisfaction
with the thermal environment and evaluating it through subjective evaluation. Furthermore, the
thermal condition of the building can affect the living quality of the occupants and the building
energy consumption to some extent (Nicol & Humphreys, 2004). Therefore, thermal comfort
assessment in the built environment is crucial to the satisfaction of occupants and building
energy consumption. Also, accurate assessment of thermal comfort of occupants is very
conducive to creating a comfortable thermal environment, improving the occupants’
satisfaction, and saving energy (Chai et al., 2020).
Predicted Mean Vote (PMV) is a common indicator used to assess the thermal comfort of
occupants, which aims to predict the mean value of votes of a group of occupants on a seven-
point thermal sensation scale (cold (−3), cool (−2), slightly Cool (−1), neutral (0), slightly
Warm (1), warm (2), neutral (0), slightly warm (2)hot (3)) through six inputs (air temperature,
mean radiant temperature, airspeed, humidity, metabolic rate and the insulation of the clothing)
(Fanger, 1970).

3

1.2 Indoor Environmental Issues
1.2.1 Indoor environment under COVID-19
The novel Coronavirus disease (COVID-19) is a highly pathogenic and contagious disease
caused by severe acute respiratory syndrome coronavirus 2 (SARS-COV-2), which emerged
in December 2019 in Wuhan, Hubei Province, China, and it spread rapidly in northern Italy
and Europe in mid-February 2020 and has been spreading globally ever since (Zoran et al.
2020). In addition, the World Health Organization (WHO) declared that COVID-19 poses a
severe threat to human health and that there is a positive correlation between air pollution and
transmission of the virus on March 11th, 2020 (Comunian et al., 2020). Specifically, in terms
of viruses, air pollution is a mixture of chemicals such as CO, NOx, O3, SOx, Particular matters
(PM), and volatile organic compounds (VOCs) that may aggravate respiratory infections
(Domingo & Rovira, 2020). Some studies have pointed out that PM2.5 (particles with
aerodynamic diameter less than 2.5 μm) is a potential airborne transmitter of SARS-COV-2,
and that droplets carrying virus particles may combine with PM to promote the spread of the
virus in the air, thus increasing the risk of human infection (Marquès & Domingo, 2022).
Moreover, the methods for ensuring good IAQ and environment, reducing the transmission rate
of air pollution and aerosol-borne viruses are worth researching.
1.2.2 Inadequate current indoor air quality standards
With the spread of COVID-19 nowadays, human exposure risk is higher indoors because
humans tend to spend a lot of time indoors. Also, the viral particles spread more easily from
person to person indoors than outdoors if the room does not have enough ventilation (Centers
for Disease Control and Prevention). Furthermore, cross-infection is the unintentional transfer
of pathogens on surfaces or between individuals (MEMIC), and cross-contamination between
individuals and the hi-tough surface is the crucial factor leading to the spread of the COVID-
19 virus. Inadequate ventilation can increase cross-infection chances of the coronavirus.
Increasing the amount of outdoor air that comes into the house can also help reduce the
concentration of indoor air pollutants or any viruses that might be in the air, helping to reduce
the risk of airborne viruses (U.S. Environmental Protection Agency | US EPA, n.d.). Thus,
providing enough ventilation to keep human health is essential.
A new study of coronavirus in buildings found that because of the virus's ability to spread and
its tendency to reside in the air over time, managers may not rely on standard rate of air
circulation in ventilation systems to remove virus particles (Sparks, 2021). Therefore, basic
ventilation and current indoor air quality standard might not help maintain human health. Also,

4

the emergence of COVID-19 has affected the lives of millions of people around the world. As
COVID-19 can be transmitted through aerosols in the air, concerns are growing about the
spread of the virus in confined Spaces, making air quality improvements urgently needed.
Moreover, ASHRAE stands for American Society of Heating, Refrigerating and Air –
Conditioning Engineers, and ASHRAE standards 62.1 and 62.2 are the recognized standards
for ventilation system design and acceptable IAQ (ASHRAE). Mechanical, electrical, and
plumbing (MEP) designers always use ASHRAE Standards as the design guideline to
determine the indoor environmental quality parameters setting and design the Heating,
Ventilation, and Air Conditioning (HVAC) system. However, different people might have
different physiological responses to the same air pollution because people have different
immune systems that are genetically influenced, and people of different genders and ages have
different immunity (Brodin & Mark, 2017). Therefore, participants might show different health
conditions in different indoor environments in residential buildings. This paper focuses on
improving indoor air quality for human health through controlled natural ventilation, helping
to build a healthier indoor environment to make more people feel comfortable and healthy; we
expect this would also be helpful during the COVID-19 pandemic.
1.2.3 Impacts of indoor air pollution on individual occupants
People built houses mainly for shelter from bad weather and security in the past. Over time,
the humble houses evolved into today's complex mechanized structures that resist climate
adversity and meet the specific requirements of indoor conditions throughout the year.
Moreover, people's requirements for indoor comfort have shifted from the basic requirement
of keeping warm or cold to the requirement of precise temperature, humidity, and air quality
conditions (Rashad et al., 2021).
However, people also become sensitive to subtle changes in temperature and humidity, and the
natural ability to adapt between weather patterns diminishes (Rashad et al., 2021). Today,
people spend most of their time indoors, increasing daily energy consumption. Moreover, with
the acceleration of urbanization, the demand for domestic energy also increases to meet
people's higher requirements for comfort. According to statistics, nearly 70% of the world's
total energy consumption is used to maintain a good indoor environment (Ganesh et al., 2021).
Therefore, it is vital to avoid wasting too much energy and maintain a balance between the
energy needs of the building and the quality of the indoor environment.
In current IEQ projects, design guidelines such as ASHRAE standards 55, 62.1, and 90.1 play
a crucial role in determining IEQ parameter settings. However, different people (such as people
of different ages, gender, health level) have different sensitivity to indoor air pollutants.
Therefore, if everyone uses the same set of design standards, it may cause discomfort for the

5

occupants. Therefore, controlling indoor air quality based on human physiological responses
is essential.
1.3 Biometric signals as indicators of human physiological states
Biometrics technology has been applied in many fields, such as in the field of identity
recognition, facial structure, voice, fingerprint, and other physiological features have been
widely used. Moreover, biometric signals can also be used as objective data to show human
health status. The biometric signals in indoor environmental quality research include skin
temperature, heart rate (HR), Electroencephalogram (EEG), and electrical activity of skin
(EDA).
Besides, more and more people use wearable technology to track their health. These wearable
technologies capture physiological signals such as heart rate, heart rate variability, and skin
temperature that are important to the body. These parameters, which objectively reflect human
physiological responsesmay act as indicators that can be used to improve personal health and
adjust indoor environmental conditions to improve comfort (Malakhatka et al., 2021).
1.3.1 Skin Temperature
As a barrier between our body's interior and external environment, our skin is the body's largest
organ, with a surface area of two square meters, accounting for about 16 percent of our body
weight, and protecting our internal environment from adverse external factors (Zimmermann,
2018). Also, our skin can promote homeostasis by sensing various disturbances that occur at
the boundary between the two environments, including thermal disturbances and triggering
defensive responses. Furthermore, skin temperature is often used to assess the temperature of
our external environment (Romanovsky et al., 2014). Thus, environmental conditions and
anthropogenic variables, such as air temperature, wind speed, and clothing, can directly affect
skin temperature. Moreover, human skin temperature has been widely used as a biometric
signal index to measure human thermal sensation (Choi & Loftness, 2012).
1.3.2 Heart Rate
Heart rate is an important indicator of human health. It measures the number of contractions or
beats of the heart per minute. Heart rate varies depending on human physical activity, safety
threats they received, and their emotional responses. While a regular heart rate is no guarantee
that a person is free of health problems, it is a valuable benchmark for identifying a range of
health problems (Heart Rate: What Is a Normal Heart Rate?, n.d.). The pulse rate is a

6

measurement of the heart rate, which also means the number of times the heart beats per minute.
In addition, the normal pulse range for a healthy adult is 60 to 100 beats per minute. Also, pulse
rates may fluctuate and increase with exercise, illness, injury, and mood. Besides, for very fit
people, such as athletes, who regularly undergo cardiovascular conditioning, may approach 40
~ 50 beats per minute without any problems (Electroencephalogram (EEG) | Johns Hopkins
Medicine, n.d.).
1.3.3 Electroencephalogram (EEG)
Electroencephalography (EEG) test detects abnormalities in brain waves or electrical activity
in the brain. During this process, electrodes made up of small metal discs with thin wires are
attached to our scalp. In addition, electrodes detect tiny electrical charges generated by the
activity of our brain cells. The charges are magnified and appear as diagrams on a computer
screen or as records that can be printed on paper (Electroencephalogram (EEG) | Johns
Hopkins Medicine, n.d.). Nowadays, various neuroimaging techniques have been applied to
assess human mental stress and measure their brain activity directly or indirectly, which include
functional near-infrared spectroscopy (fNIRS), EEG (Al-Shargie et al., 2015), and functional
magnetic resonance imaging (fMRI) (Zhang et al., 2019).
1.3.4 Electrical activity of skin (EDA)
Electrical skin activity (EDA, sometimes called galvanic skin response) is a change in skin
conductance due to sweat secretion. The data is collected by applying a low, undetectable
constant voltage to the skin and then measuring how the skin conductance changes. Also, EDA
signals can reflect the intensity of our emotional state (also known as emotional arousal). Our
level of emotional arousal varies depending on the environment we are in and the events we
are faced with, such as scary, happy, or other emotion-related events, and the subsequent
changes in our emotional response to that event can also increase skin conductance (Farnsworth,
2019).
Besides, Skin conductance response (SCR), one component of the galvanic skin response (GSR)
activity, is proportional to the number of activated sweat glands, and essentially, the more
agitated a person is, the more SCR increases. In addition, SCR is often regarded as the "peak"
of activity (GSR peak) because it represents rapid signal values. If an SCR occurs in response
to a stimulus (usually within 1-5 seconds), it is called an Event-Related SCR (ER-SCR), and if
it occurs without any discernible cause, it is called a Non-Specific SCR (NS-SCR) (Farnsworth,
2019).

7

Figure 1. 1 Illustration of SCR, SCL and EDA Peaks. (Retrieved from https://imotions.com/blog/skin-conductance-
response/)
EDA is one of the most basic indicators of the state of the autonomic nervous system. EDA is
controlled by the sympathetic nervous system, which is different from another indicator, heart
rate, which is influenced by both the sympathetic and parasympathetic nervous systems
(Sarchiapone et al., 2018). We can observe increases in sympathetic activation by monitoring
subtle electrical changes across the skin's surface. The sympathetic nervous system prepares
our body to take action during the time of threat, stress, or excitement. For example, when
people feel nervous and suddenly their heart starts beating. These are all signs that the
sympathetic nervous system is taking over and preparing to respond. Therefore, while we may
not be in immediate danger, the system has evolved to facilitate an immediate response to stress
or threat, and our heart rate increases simultaneously (Understanding the Stress Response -
Harvard Health, n.d.).
1.4 Machine Learning techniques
Machine learning (ML), a branch of artificial intelligence, is a collection of algorithms and
techniques used to design systems that learn from data, enabling machines to improve

8

empirically at a given task (Lee, 2019). In this study, the data on indoor climate and air quality
will be collected from the participants. Furthermore, Machine learning (ML) is being
extensively used to predict air quality thanks to its prediction performance (Gunjal &
Kamalapurkar, 2020.). Thus, machine learning will be applied to build the prediction model
for analyzing the data.
1.4.1 Linear Regression
Linear Regression uses continuous variables to predict the actual values, which is also one of
the simplest algorithms to model the relationship between the dependent variable and
independent variables by using a straight line (Lee, 2019). Linear regression analysis could be
used to determine the relationship between various factors (air quality components and the
appropriate environmental parameters) and IAQ, which can be considered a baseline method.
1.4.2 Random Forest
Random forest is a commonly used machine learning algorithm, which is composed of multiple
decision trees and combines the outputs of multiple decision trees to achieve a single result
(van der Heide et al., 2019), which can deal with classification and regression problems with
high accuracy (Ouedraogo et al., 2019). One of the most important characteristics of the
random forest algorithm is that it can deal with data sets containing continuous variables, such
as the case of regression and classification variables. Bootstrap Aggregation is the working
principle and integration technology used by random forest, which selects a random sample
from the collected data set. Therefore, the models created by random forest are all generated
from samples provided by the original data set, and each model is independently trained. The
final output is based on average or majority ordering (using the average answers given by a
large number of trees), with high stability (Sruthi, 2021).
This study will use Linear Regression and Random Forest for data analysis and model creation.
1.5 Summary
The first part of the first chapter describes the importance of IAQ, indoor environmental quality,
indoor ventilation, and human thermal comfort. It also raises the issue of indoor air quality and
environmental quality, and its impact on different people, especially in the context of COVID-
19. The second part of this chapter introduces some physiological signals that objectively
reflect human health. Finally, the last part introduces some technical methods and algorithm
models of Machine Learning.

9

Chapter 2: LITERATURE REVIEW

For understanding the impact of IAQ on human health, knowing the correlation between IAQ
parameters (especially CO2 and PM2.5) and human physiological responses, and learning how
to maintain good air quality, many studies have been reviewed in this chapter.
2.1 Impact of indoor air pollution on human being
2.1.1 Importance of maintaining good indoor air quality
IAQ is critical to our human health, and it is estimated that people spend more than 90% of
their time indoors in general although this varies across contexts (Chang & Gershwin, 2004).
The Disability Adjusted Life Years (DALY) indicator is regarded as “loss of life corrected by
disability,” which is widely used by the World Health Organization (WHO) to measure a
disease burden on the population and identify the causes of disease. Wysocka used DALY to
evaluate the health effects of air pollution on the human body. The most common pollutants
with the most significant impact on air quality deterioration include carbon monoxide and
dioxide, nitrogen oxides, volatile organic compounds (VOC), formaldehyde, radon, tobacco
smoke, particulate matter, and microorganisms in the air, which have a significant impact on
air quality deterioration. For example, moderate to high levels of CO2 can cause headaches and
fatigue, while higher levels can cause nausea, dizziness, and vomiting. Among the above types
of pollution sources, PM2.5 pollution causes 78% of the disease burden, accounting for the
highest proportion, followed by radon (8%) and air humidity (5%) (Wysocka, 2018).

Figure 2. 1 Burden of disease from poor IAQ by pollutant - EU26 - 2010. (Retrieved from https://www.e3s-
conferences.org/articles/e3sconf/abs/2018/24/e3sconf_solina2018_00133/e3sconf_solina2018_00133.html)

10

Fig.2.2 shows the total disease burden resulting from poor air quality in 26 European countries.
More than half of the cases (55%) were cardiovascular diseases, and this was followed by lung
cancer (22 %). In addition, other diseases associated with poor indoor air quality include
asthma and allergies (12%), chronic obstructive pulmonary disease (7%) and respiratory
infections (3%) (Wysocka, 2018). Therefore, it is crucial to maintain good IAQ.

Figure 2. 2 Total burden of disease - EU26 - 2010. (Retrieved from https://www.e3s-
conferences.org/articles/e3sconf/abs/2018/24/e3sconf_solina2018_00133/e3sconf_solina2018_00133.html)
Moreover, some airborne virus or bacteria can cause disease, and small outbreaks can occur
inside buildings (Chang & Gershwin, 2004), such as SARS and COVID-19, which are airborne
viruses. Therefore, people pay more and more attention to indoor air quality, and because of
the advent of the Covid-19, indoor air quality has become the focus of widespread concern
(Starr, 2021).
2.1.2 Impacts of indoor air quality on work efficiency
As a result of the energy crisis of the 1970s, buildings in the United States were built as airtight
and energy efficient as possible, and indoor air quality was not a priority. As a result, toxic
volatile organic compounds (VOCs) and exhaled carbon dioxide accumulated in indoor air
quality (Allen et al., 2016). In the study of Allen and his team, they tested 24 white-collar
volunteers on their analytical thinking and ability to cope with a crisis for six consecutive days
and examined the effects of IAQ on the test results by changing the indoor ventilation rates,
CO2 and VOC levels. The results showed that when the volunteers worked in well-ventilated
conditions (with low levels of CO2 and VOC), their scores were 61% higher than when they
worked in a typical office building (Allen et al., 2016).

11

2.1.3 Important factors of indoor air quality under the Covid-19
2.1.3.1 PM 2.5 as an important indicator
In daily life, we breathe airborne particulate matter (PM) to some extent, but most of them are
easily removed by mucus cilia in our respiratory system (Pozzi et al., 2003). However, PM2.5
can not only penetrate the gas-exchange zone in the lungs, but it can also pass through the
body's respiratory barrier into the circulatory system, and can even spread throughout the body
(Xu et al., 2008). In addition, PM2.5's large surface area makes it more likely to bind to toxic
compounds (Pandey et al., 2013). Therefore, PM2.5 is more correlated to the adverse effects
on human health than larger particulate matter (Feng et al., 2016). Furthermore, the WHO Air
Quality Guidelines recommended PM2.5 as an indicator of air pollution particles, which
gradually gained people's attention.
Four researchers, Silvia Comunian, Dario Dongo, Chiara Milani, and Paola Palestini found
that the daily concentration of Particulate Matters (PM) in Italian cities during the COVID-19
outbreak was higher than the permitted annual concentration in the months prior to the outbreak.
They concluded that the relationship between the spread of COVID-19 and air pollution is
positive. Furthermore, atmospheric PM plays a critical role in viral transmission, morbidity,
and mortality, and exposure to PM will increase the severity of symptoms in patients with
COVID-19 (Comunian et al., 2020). Maria A Zoran, Roxana S Savastru, Dan M Savastru,
and Marina N Tautan studied the relationship between the degree of accelerated diffusion and
lethality of COVID-19 and the surface air pollution (PM2.5, PM10) in the Milan metropolitan
area in Italy (Zoran et al., 2020).
Their results demonstrated that the daily averaged ground levels of particulate matter
concentrations strongly influence the accelerated extent of the spread of COVID-19 and the
increase in cases in Milan (Zoran et al., 2020). In addition, Xiao Wu, Rachel C.
Nethery, Benjamin M. Sabath, Danielle Braun, and Francesca Dominici collected COVID-19
deaths from more than 3,000 counties (98% of the population) in the United States from the
Systems Science and Engineering Coronavirus Resource Center at Johns Hopkins University.
These five researchers explored whether long-term average exposure to fine PM2.5 would
enhance the risk of COVID-19 death in the United States. They found that a mere 1 mg/m3
increase in PM2.5 was associated with an 8% increase in COVID-19 mortality. Therefore, they
believed that a slight increase in long-term exposure to PM2.5 could significantly increase the
COVID-19 death rate and co-exposure to PM2.5 air pollution results in greater vulnerability
and risk (X. Wu et al., 2020).

12

2.1.3.2 CO2 as an important factor
As a carrier to release infectious particles in the air, the amount of CO2 in exhaled air is 40,000
PPM, much higher than in outdoor air (350ppm). Since CO2 is also exhaled and carried in the
air circulation (Bhagat et al., 2020), and because most buildings have no significant internal
source of CO2 other than that from the occupants themselves. Also, when there are no other
sources of indoor pollution, human respiration is the primary source of indoor carbon dioxide
pollution (Chen & Hsiao, 2015). Therefore, CO2 can be regarded as a substitute for exhaled air
(Rudnick & Milton, 2003). In addition, since airborne infectious diseases can only be obtained
by inhaling previously exhaled air, and continuous monitoring of CO2 concentration is
relatively cheap and convenient, which could avoid the necessity of directly measuring the
outdoor supply rate. Four researchers, Rajesh K. Bhagat, M. S. Davies Wykes, Stuart B.
Dalziel, and P. F. Linden believed that CO2 levels could be used to indicate the possible
presence of SARSCOV-2 in the air. Also, they found that when CO2 levels were above 775
ppm, which is a sign of insufficient ventilation and a need for supplemental measures to reduce
the risk of infection. In addition, they found that the risk of infection increased with exposure
time, and the CO2 levels increased over time when people occupied a space. Hence, the number
of people in the room and how long people occupy the room are essential parameters for
measuring IEQ (Bhagat et al., 2020).
2.2 Impact of indoor air pollution on individual occupants
According to the 2019 report "Life and breath: How air pollution affects health in Minnesota,"
this report analyzed 2013 county-level health data on air quality from MPCA and Minnesota
Department of Health (MDH), assessed the impact of air pollution on human health in each of
Minnesota's 87 counties, and analyzed vulnerability to air pollution concerning demographic
factors such as age, race, and poverty. The study found that vulnerable groups such as the
elderly, people with pre-existing heart or lung conditions, and children with asthma are
disproportionately affected by air pollution. In addition, the health effects of air pollution are
not just a problem in large cities. Statewide, the health effects of air pollution are higher in
counties with older populations, or those with no health insurance or high levels of poverty
(Life and Breath Report).
2.3 Machine learning in building science domain
Machine learning algorithms are often used to solve complex nonlinear problems because of
their excellent self-learning ability and ability to find optimal solutions quickly. Machine
learning algorithms have also been applied in several thermal comfort studies to evaluate the

13

thermal comfort of occupants (Chai et al., 2020). Four researchers, Chai Qian, Huiqin Wang,
Yongchao Zhai, and Liu Yang, used two machine learning models, SVM and Artificial neural
networks, to predict the thermal comfort votes (TCV) and thermal sensation votes (TSV) of
residents in naturally ventilated residential buildings. They found that the machine learning
(ML) model can quickly analyze the relationship between input and output parameters. In
addition, traditional PMV was developed by Fanger based on many experimental tests.
Compared with PMV, the machine learning model performed better in predicting TCV and
TSV of residents in naturally ventilated residential buildings with higher accuracy (Chai et al.,
2020). This study applied Stepwise Linear Regression and Random Forest to analyze the data
collected, including indoor climate data (temperature, relative humidity, CO2, PM2.5, acoustic
and lighting), human physiological responses (heart rate, stress level, EDA, and skin
temperature), outdoor PM2.5 and survey data.
2.4 Indoor environmental quality and human physiological responses
2.4.1 Human physiological responses and bio-signals in building science domain
Indoor environmental quality will affect the health of the occupants. Besides, the physical state
of the occupants can be known by analyzing their physiological responses (Bonato, 2005).
Physiological responses include heart rate, respiration, skin temperature, and perspiration. For
example, changes in sweat are measured by skin electrical response measurements to detect
changes in electrical conductivity (Bergstrom & Schall, 2014). In Okamoto-Mizuno and
Mizuno's study, they found that physiological parameters such as skin temperature and heart
rate variability are related to the thermal comfort of waking people (Okamoto-Mizuno &
Mizuno, 2012).
Sun, Yu Ping, and Neng Zhu analyzed many relevant references on the human body's
physiological response to humid and hot environments. They found that high temperature and
humidity in the working environment significantly impacted human health. High temperature
and humidity would reduce the temperature difference between inside and outside the human
body, resulting in metabolic difficulties and physiological reactions such as heart failure and
hypoxia (Sun & Zhu, 2013). In addition, Höppe et al. found that subjects' skin humidity
increased when the temperature increased (Höppe et al., 2000).

2.4.2 The correlation between indoor air quality and human physiological responses
Two main factors affect human physiological responses in the indoor building environment:

14

IAQ and indoor climate factors. According to ASHRAE Standard 62.1, IAQ has a considerable
influence on the physiological response of occupants, and CO2 concentration is one of the main
factors. Besides, ASHRAE recommends keeping indoor CO2 below 1000 PPM. As for the IAQ
factors, Kajtar and Levente found that increased CO2 concentration increased heart rate and
diastolic blood pressure (DBP) (Kajtár & Herczeg, n.d.). Besides, Chen, Ya-Chan, and Tzu-
Chien Hsiao simulated poor ventilation conditions in a closed room to explore the effects of
ventilation in poor rooms with different CO2 levels on human physiological responses, and the
research results showed that high CO2 concentration would lead to elevated skin temperature
(Chen & Hsiao, 2015).
Researchers Liu et al investigated the response of human cardiovascular system during exercise
under different concentrations of CO2 in oxygen-rich rooms at high altitude, and found that the
heart rate of subjects increased significantly under altitudes of 3800 meters and higher
concentrations of CO2 (5%) ((Liu et al., 2015). In addition, researchers Bailey et al. studied
the effects of relative higher CO2 concentration (7.5%) on 20 participants' physiological
responses, and each participant stayed in a closed room (the air in the room was mixed with
7.5% CO2) for more than 20 minutes. Based on the objective data from cardiovascular
measures and participants' subjective ratings, the researchers concluded that the air with 7.5
percent CO2 increased the experimenters' heart rate and anxiety, and decreased the participants'
feelings of relaxation and happiness, compared with air. In addition, at the same time, the
subjects' fear and other subjective symptoms associated with anxious states also increased
(Bailey et al., 2005). Besides, Seddon et al. found that anxiety responses in patients with
generalized anxiety disorder increased after 20 minutes of 7.5% CO2 inhalation (Seddon et al.,
2011).
Vilcassim et al found that compared with low pollution cities (PM2.5 < 100 μg/m
3
), exposure
to cities with high levels of pollution (PM2.5 > 100 μg/m 3) caused small but statistically
significant acute changes in cardiopulmonary function and respiratory symptoms in healthy
young adults visiting the cities. Their results showed that traveling to a foreign city with high
levels of air pollution can adversely affect one's cardiorespiratory health (Vilcassim et al.,
2019). Furthermore, a study of Vallejo et al showed that an increase in the ambient
concentration of particulate matter (PM) [sub.2.5] was associated with a decrease in Reduced
heart rate variability (HRV) in healthy residents (Vallejo et al., 2006).
HRV has been reported to be associated with increased mortality and morbidity from
cardiovascular disease. Previous studies of Creason et al and Holguin et al found that ambient
PM2.5 was associated with short-term HRV reduction in the elderly (Creason et al., 2001;
Holguín et al., 2003). Moreover, in the study of Paoin et al, they found that short-term exposure
to PM2.5 was associated with lower 24-H HRV in patients with symptoms/signs of
heart diseases (Paoin et al., 2020). However, some previous studies reported no significant

15

association between PM2.5 exposure and HRV (Suh & Zanobetti, 2010; C. F. Wu et al., 2010).
For verifying the results from the previous study, our study identified the correlation between
two IAQ parameters (indoor CO2 concentration and indoor PM2.5 concentration) and human
physiological responses (heart rate, stress level); and the results of our experiment showed
different results. Besides, little research has documented the impacts of PM2.5 and CO2
concentrations on EDA and human skin temperature. Therefore, our study also researched the
correlation between two IAQ parameters and those two bio-signals.
2.5 Effective control/Energy efficient control
2.5.1 Improve indoor air quality/overcome the air pollution
IAQ can be improved in many ways, such as providing proper ventilation, or using engineering
controls such as air filtration. In addition, indoor air filtration can be provided through an
HVAC system, a portable indoor air purifier, or an integrated system through whole-house
filtration.
a. Ventilation
Since people spend 80 to 90 percent of their time in indoor Spaces, diluting pollutants in those
Spaces is a primary factor in improving indoor air quality and maintaining a healthy
environment (Morawska et al., 2020). Ventilation can dilute air pollutants in indoor space by
exchanging indoor and outdoor air, and natural ventilation can be achieved through windows
and vents. In addition, HVAC systems can mechanically provide ventilation and can be
engineered to reduce the likelihood of increased infection. However, because HVAC systems
recirculate air, taking more energy to bring in the fresh air, recirculation could increase
infection rates if the system does not have proper filtration installed (Agarwal et al., 2021).
According to ASHRAE and REHVA guidelines, air recirculation in indoor environments,
especially centralized air conditioning systems, should be banned in the context of the current
COVID-19 epidemic.
b. Smart control
When residents feel uncomfortable in a building, they will most likely adopt adaptive behaviors
to reduce discomfort and meet their comfort requirements, such as adjusting windows and
using fans or HVAC systems. However, everyone's residential behavior is different, and
individual differences can lead to differences in the indoor environment and energy use in
residential buildings (Calì et al., 2016). Junseok Park, Bongchan Jeong, Young-Tae Chae,

16

and Jae-Weon Jeong studied the occupants' behavior of the manual control of windows for
cross-ventilation and obtained field monitoring data from 23 samples homes. (a. Outdoor
environment: air temperature and relative humidity; b. Indoor environment: air temperature,
relative humidity, CO2 concentration, and mean radiation temperature; c. Window status:
opened or closed), the positions of various detectors are shown in Figure 2.3.

Figure 2. 3 The floor plan of the typical type of sample homes showing the location of humidity, temperature and CO2 data
logger (Lutron data loggers, MCH-383SD) and binary status logger (UX90-001, Onset Computer) (Retrieved from
https://journals-sagepub-com.libproxy2.usc.edu/doi/full/10.1177/1420326X20927070 ).
In addition, the authors collected outdoor weather data (solar radiation, wind speed, and PM10
concentration) affecting the window state from local weather stations. After using computer
algorithms, the authors found that although the outdoor temperature was an essential factor for
occupants to open or close their windows, it was difficult to use it alone to predict window
opening or closing behavior. However, when indoor environmental factors were considered
with outdoor temperature, window opening and closing behavior were well predicted (Park et
al., 2020).
2.5.3 Energy saving
William J.N. Turner and Iain S. Walker presented a new hybrid ventilation system that uses
intelligent ventilation remote control to minimize energy consumption. The authors used
computer simulations to study the effects of passive and mixed ventilation systems on energy
and indoor air quality (IAQ) in 16 climate zones in California. They found that controlled
passive and mixed ventilation can provide indoor air quality comparable to continuous
mechanical ventilation and has lower energy consumption (Turner & Walker, 2013).
Rashad et al.'s research review paper advocates using available environmental energy in homes
to improve human thermal comfort and reduce energy consumption. Also, they reviewed five

17

passive energy-saving strategies: Passive architectural design, Night Ventilation (NV),
Nocturnal Radiation Exchange (NRE), Phase Change Materials (PCM), and Indirect
Evaporative Cooling (IEC). Although these energy-saving strategies can help energy saving to
a certain extent, they need the relevant departments to vigorously promote energy-saving and
enhance the execution force to implement them better. Meanwhile, more efficient HVAC
systems need to be further studied, which greatly benefits to energy saving (Rashad et al., 2021).
2.6 Summary
The second chapter mainly reviews previous studies investigating the impacts of IAQ on
people's lives and health, and the impacts of IAQ parameters (PM2.5 and CO 2) on human
physiological responses. In addition, this chapter explains the significance of PM2.5 and CO2
as IAQ parameters. Also, this chapter reviews some important research of Machine Learning
in the field of building science and human physiological response and points out the importance
and accuracy of multi-machine learning in data analysis. Finally, some methods for controlling
indoor air quality were summarized at the end of the second chapter.
Although many studies have researched the effects of indoor air on human health, few studies
focused on which human bio-signals are most affected by IAQ, and few have tried to control
IAQ based on human physiological responses. Therefore, this study focused on the impacts of
IAQ on four kinds of human physiological responses (heart rate, stress level, skin temperature,
and EDA) and explored the correlation between each physiological signal and IAQ, and finally
determined the most relevant bio-signals and factors to IAQ Evaluation.
Chapter 3: METHODOLOGY
3.1 Project Design
The research goal is identifying the relationship between indoor air quality (PM2.5 and CO2)
and human physiological responses, investigating the most relevant bio-signal to the IAQ
parameters (PM2.5 and CO2) based on the analysis results, and exploring the more relevant
bio-signals and other factors (including indoor temperature, indoor humidity, IAQ parameters)
to IAQ Evaluation and thermal comfort of each participant and finding the common features
across all participants.
In addition, based on the results from this study, our long-term goal concludes develop the IAQ

18

control model based on the relevant bio-signals, operate, and apply the model in a real building.
But these two progresses belong to our future work, and this study only focus on first three
procedures showing in Fig 3.1.

Figure 3. 1Whole progress
In this research, it is necessary to collect enough and relatively accurate data before using a
machine learning model and statistical tools to analyze data. Three types of data need to be
collected: indoor and outdoor environmental quality conditions, participants' bio-signals, and
surveys. Firstly, the indoor environmental data include the data of air temperature, humidity,
PM 2.5 concentration, CO2 concentration, acoustic, and lighting. Besides, air temperature,
humidity, PM 2.5 concentration, CO2 concentration are the outdoor environmental parameters
measured in this study. Secondly, the occupants' bio-signal data include heart rate, skin
temperature, stress level, and EDA. Finally, two surveys collected participants’ perceptions of
indoor air quality and thermal comfort.
12 participants (six males and six females) from USC (University of Southern California)
School of Architecture volunteered to participate in this study (this study ID on IRB is UP-21-
00874), their range in age from 24 ~ 35. All the participants were healthy and had no history
of asthma. Also, the experiment took place in the room where they spent the most time at home.
Each one participated for five consecutive days. The sensors collecting indoor and outdoor
environmental quality parameters were installed in the subjects' rooms one day before the
experiment began. In addition, the sensors for measuring outdoor air quality data were placed
on the experimenter's outdoor balcony one day before the experiment began. Besides, two
smartwatches, which measured human physiological responses, were calibrated the day before
Investigate the
correlation
Build the model
(perception/sensation)
Determine the most
relevant bio-signals
Develop the IAQ
control model based on
the relevant bio-signals
(command)
Actuator and operation

19

the experiment. The participants had worn the two smartwatches for an hour to see if there was
any data collected to ensure the experiment could be carried out properly. Also, two
questionnaires (one was the assessment of IAQ and the other one was thermal comfort) were
sent to the subjects in word form, and the subject filled in the data every two hours during the
five–day experiment when they were in their room (not including sleep time). The experiment
lasted from 0:00 AM (GMT-8) on the first day to 0:00 AM ((GMT-8) on the fifth day, and the
time interval for data collection for all sensors was 10 minutes (except for the illuminance
sensor).
All experimenters provided written informed consent before their participation, and each
participant got a $50 Gift Card after finishing the experiment.
In addition, the statistical tools (SPSS and Minitab) and machine learning would be applied to
analyze these data to identify some relationships, including the correlation between IAQ
parameters and human bio-signals, the relationship between IAQ Evaluation and human bio-
signals, and the relationship between thermal comfort and bio-signals. After analyzing all the
participants' data collected, common features of physiological responses across all the
participants in their indoor environmental quality condition were found, and most relevant bio-
signals were determined.

Figure 3. 2 Methodology diagram
3.2 Hardware and Software
Multiple sensors were used to record indoor air measurements and occupants' physiological

20

responses. Moreover, the data analysis helped to explore the correlation between human bio-
signals and indoor environmental parameters, especially IAQ parameters (PM2.5 and CO2). 12
participants participated in this study and were required to wear two biometric sensors (Garmin
Vivosmart3 and Empatica Embrace 2) when they were in their bedroom where the sensors were
placed. The sensors collecting indoor climate data were installed in the participants’ bedrooms,
and the sensors collecting outdoor climate data were installed outside their apartments.
3.2.1 Sensors for IEQ measurements
IEQ data objectively reflect the current state of the indoor environment, including lighting,
acoustic, air temperature PM2.5 concentration, and CO2 levels. Based on ASHRAE standards
55, 62.1, and 90.1, the recommended range for these indoor environments' parameters’ values
are shown below:
Table 3. 1 IEQ guidelines
Parameters Guideline
Workstations Illuminance level (lux) 200~500 lux
Acoustic decibel (dBA) <40 dBA
Indoor temperature (℃) 19~28℃
PM2.5 <25 μg/m
3
CO2 Level (ppm) <1000ppm
3.2.1.1 Sensor for collecting air temperature and CO2 concentration
Before starting the experiment, several appropriate sensors for the experiment were selected.
First, HOBO MX 1102 was selected to collect indoor temperature, relative humidity, and CO2
concentration levels. Because this sensor has a USB port, it can be connected to a computer
running HOBOware to set the parameters for measuring data, and then we can read data,
download the data in excel format. In addition, the data collection interval was ten minutes,
and all collected data were automatically stored in memory. The time interval for measuring
CO2 was ten minutes.

Figure 3. 3 HOBO MX1102

21

Table 3. 2 HOBO MX1102 Details
Technical details

size 7.62 x 12.95 x 4.78 cm
Weight 267.4 g
Temperature Sensor

Range 0℃~50℃
Accuracy ±0.21°C from 0° to 50°C
Resolution 0.024℃ at 25℃
Response time 12 mins to 90% in airflow of 1 m/s
CO2 sensor

Range 0-5,000 ppm
Accuracy ±50 ppm ±5% of reading at 25°C
Calibration auto or manual to 400 ppm

3.2.1.2 Sensor for collecting Particulate Matters concentration
For collecting the data of PM 2.5, one air quality sensor called Pa- II- SD manufactured by
Purple Air company was selected; it counts particles in certain size ranges and then it uses a
calibration equation to estimate PM2.5 mass concentration for residential, commercial, or
industrial use in real-time. In addition, built-in Wi-Fi enables this air quality detector to
transmit data to a PurpleAir map and store data on the map for use by any smart device.
Furthermore, this sensor can integrate an SD card and a real-time clock for locations with
limited or no Wi-Fi access, allowing itself to record and store data locally. The time interval
for measuring data for this study was ten minutes.

Figure 3. 4 Pa- II- SD

22

Table 3. 3 Pa- II- SD Details
size 85 mm x 85 mm x 125 mm
Weight 357g
Laser Particle Counters

Range of measurement 0.3, 0.5, 1.0, 2.5, 5.0, & 10 μm
Counting Efficiency 50% at 0.3μm & 98% at ≥0.5μm
Effective range (PM2.5 standard) 0 to 500 μg/m³
Maximum range (PM2.5 standard) ≥1000 μg/m³
Maximum consistency error
(PM2.5 standard)
±10% at 100 to 500μg/m³ &
±10μg/m³ at 0 to 100μg/m³
3.2.1.3 Sensor for collecting lighting
Dr.Meter LX1330B Digital Illuminance Light Meter was applied for measuring the indoor
illuminance data. However, it does not have a storing function. In this study, the illuminance
data will be measured several times in one day by the participant (morning, afternoon, night).

Figure 3.5 Dr.meter LX1330B Digital Illuminance Light Meter
Table 3. 4 Empatica Embrace 2 Details
Display 3-1/2 digit 18mm LCD
Ranges: 0.1-200/2,000/20,000/200,000 Lux
Accuracy ±3% ±10 digits (0-20,000 lux) / ±5% ±10 digits (over 20,000 Lux)
Repeatability ±2%
Sampling rate 2-3 times per second
3.2.2.4 Sensor for collecting acoustic
In terms of acoustic data collection, PCE-SDL 1 was used to measure sound level, and this
sound level data logger is designed for noise, quality control,, and all kinds of environmental

23

sounds measurement. This study is about indoor environmental quality, so this device is
appropriate. The time interval for measuring acoustic was ten minutes.

Figure 3. 6 PCE-SDL 1
Table 3. 5 Empatica Embrace 2 Details
Dimension 130×30×25mm
Weight 20g
Measure range 30dB-130 dB
Standard Applied IEC 61672-1 (CLASS 2)
Accuracy ± 1.4 dB
Frequency range 31.5Hz to 8kHz
Resolution 0.1 dB
3.2.2 Sensors for collecting human bio-signals
3.2.2.1 Sensor for collecting heart rate and stress level
Garmin Vivosmart 3 was used to collect real-time data of human physiological responses,
including the experimenter's heart rate and stress level. The time interval between two data
collections was 10 minutes.

Figure 3. 7 Garmin Vivosmart 3
Table 3. 6 Garmin Vivosmart 3 Details

24

Measured parameter Heart rate; Stress level
Range 0~200 bpm; 0~100
Resolution 1bpm; 1 unit of stress level
3.2.2.2 Sensor for collecting skin temperature and EDA
Empatica Embrace 2 was used to collect real-time data of the experimenter's electrodermal
activity (EDA). The time interval of two data collection is also 10 minutes, consistent with
other sensors.

Figure 3. 8 Empatica Embrace 2
Table 3. 7 Empatica Embrace 2 Details
Measured parameter Skin temperature; EDA
Range -40 to 115°C; 0.01 μSiemens to 100 μSiemens
Resolution 0.02℃; 1 digit~900pSiemens
Accuracy ±0.2℃ within 36~39℃; /
3.2.3 Survey
The research studies the indoor air quality control for human health in residential buildings,
and the experiment's location we chose is the experimenters' bedroom; the reason was that the
occupants would spend more time there, like sleeping and studying.
Two surveys will be used to understand the occupants' perception of indoor environmental
quality. The first one is the occupants' feeling about the indoor air quality. In addition, because
the temperature and humidity will be measured in this study, which is related to human thermal
comfort, the survey asking about human thermal comfort will also be used in this experiment.
These two surveys will distribute to each experimenter online. Besides, before their experiment,
they were asked to fill out their age and gender first. Furthermore, a five-point scale will be
used to evaluate the occupant's feelings about indoor air quality to improve the accuracy of the
evaluation, and a seven-point scale will be used to evaluate the occupants' thermal comfort to

25

improve the accuracy of the evaluation. Figures 3.8 and 3.9 show the surveys used in the
experiment.

Figure 3. 9 Indoor air quality Evaluation Survey

26

Figure 3. 10 Indoor thermal comfort survey
Python will be applied for data cleaning, data organization, data analysis, machine learning,
creation of prediction model.

27

3.3 Data collection
3.3.1 Participants’ personal information data
The participants' personal information data mainly included age, gender, and the metabolic data
included body mass index. These factors all require users to enter data before or during testing
actively. This study did not collect the participants' names for protecting the participants'
information. And this experiment has received the Institutional Review Board (IRB) approval
(study ID is UP-21-00874).
According to the previous discussion in the first chapter, people of different ages, genders,
physical and health conditions have different sensitivities to indoor ambient air quality, but the
degree of influence is uncertain. Besides, these user background data were input in the
questionnaire by the experimenter before the experiment. Moreover, in the data analysis
process, regression analysis could be carried out simultaneously with the recorded data of
indoor environmental quality parameters and questionnaire data to establish a more accurate
personal indoor air quality and thermal comfort preference curve. It is worth noting that all
measurements in this study were done in as-is condition, without controlling anything.
3.3.2 Participants’ physiological responses data
Garmin Vivosmart3 is designed to measure heart rate and stress level, and this Smartwatch can
store data that can be used for more than five days on a full charge. Before the test, the
participants downloaded “Garmin Connect” software on their mobile phones. Then, when
synchronizing data, they enabled the Bluetooth on their phone to connect to Smartwatch.
Meanwhile, the data could be automatically uploaded to the mobile phone and the official
website database. However, the data downloaded from the official website were FIT files rather
than Excel or CSV formats, so Python was needed to convert FIT files into CSV files so that
the data desired by this study could be read.
Embrace 2 is primarily used to measure skin temperature and EDA, and it can also store data.
Before the test, users installed the software "Mate for Embrace" on their mobile phones to
match the watch and establish a connection to synchronize data and upload it to the official
website database. However, unlike the Garmin Vivosmart 3, it needs to be charged every two
days, and the "Mate for Embrace software always had problems connecting to the smartwatch.
Besides, EDA and skin temperature data recorded and stored by Embrace 2 could be accessed
and downloaded for a $600 monthly subscription on Embrace Research's website. Because the
data downloaded from Embrace Research was huge because the time was calculated in

28

milliseconds, it also needed a Python script to turn it into data that can be read.
Although EEG is also an important bio-signals data for human health, this study did not
measure EEG data from all the participants. The reason was that the sensor measuring EEG
needed to be worn on the head, and if the occupants worn the device for more than around two
hours, their head would feel uncomfortable. In addition, our research needs each participant to
wear the device for five days when they are in their room, which is unrealistic.
3.3.3 Environmental parameters data
Indoor and outdoor environmental data mainly include indoor and outdoor air quality (PM2.5
and CO2), indoor ambient temperature, indoor ambient relative humidity, indoor sound level,
and indoor lighting. Two HOBO sensors were used to collect indoor and outdoor CO2, indoor
temperature, and humidity data. Participants would place the HOBO MX CO2 logger used for
indoor measurement near their work and rest place, and the HOBO MX CO2 logger used for
outdoor measurement would be placed on the balcony of participants. “HOBOware” software
needs to be downloaded on the computer before the test, which is used to set parameters (such
as time interval and test parameters) before the test. After the test, the environment data can be
directly downloaded from the computer. In addition, the data file is in CSV format, and no
further operation is required.
Two PA-II-SD sensors were used to collect indoor and outdoor PM2.5 data. In addition,
participants needed to place the PA-II-SD sensors used for indoor measurement close to their
work and rest positions. PA-II-SD, for outdoor use, will be placed on the balcony. Furthermore,
PA-II-SD sensors need to be plugged in and connected to the WIFI in the room before use and
then registered on the official website for use. The PM2.5 data can be downloaded directly
from the official website. The data file is in CSV format, and no further processing is required.
For real-time indoor acoustic data collection, one PCE-SDL 1 sound logger was applied in this
study. “Sound Datalogger” software needed to be downloaded on the computer to set up the
parameters of the logger, such as sampling rate and LED Flash Cycle. The acoustic data
recorded can be downloaded from the software, and the files downloaded are CSV files, which
do not need further data processing.
In terms of indoor lighting measurement, Dr.Meter LX1330B Digital Illuminance Light Meter
was used, but this sensor. This detector was the only one that could not store data and could
only read real-time data in this study. Besides, each experimenter selected one day and recorded
real-time data of ambient illuminance as well as maximum and minimum values three times
(morning, noon, and evening).

29

3.3.4 Survey data
It is challenging to collect user feedback because this study requires participants to participate
for five consecutive days, and users need to fill out the surveys every two hours. Sometimes
the participants would forget to fill the survey in time, so the collection of questionnaire data
mainly depends on users' awareness and reminders from researchers. First, each participant
received two Word documents, roughly every two to three hours (during non-sleep periods);
participants needed to fill out two surveys on thermal comfort (-3, -2, -1, 0, 1, 2, 3) and their
perception of air quality (-2, -1, 0, 1, 2).

Figure 3. 11 All data collected from the first participant
3.4 Machine learning
3.4.1 Data Organization
After data collection, all the data for each participant needed to be organized in one excel file,
and each recorded data must correspond to the same time. For example, the heart rate data at
2:10 PM on November 14 should correspond to the PM2.5 data at 2:10 PM on November 14
th
.
3.4.2 Data Analysis
Two software, Minitab and Python, were applied in this study. After organizing the data, the
data collected for each occupant were input in Minitab and Python to analyze the data further.
Firstly, stepwise linear regression was applied to identify the correlation between indoor air

30

quality and human physiological responses and determine the correlation through Minitab.
Secondly, random forest was used to checking the prediction model’s accuracy, which used
the most correlated human physiological signals to predict the evaluation of indoor air quality.
In the data collection process, as the participants will not stay all the time indoors, the data of
human physiological signals were missed when the participants were not indoors, so the
missing data should be cleaned before data analysis.
3.5 Summary
The third chapter mainly describes each step and the overall framework of the research,
including the selection of sensors, the process of data collection, and the method of data
analysis after collection. Finally, the fourth chapter will detail the data analysis, the process of
establishing the prediction model, and the analysis of the results.

31

Chapter 4: DATA ANALYSIS
In most cases, the Pearson correlation coefficient (also known as the Pearson product moment
correlation coefficient) r, which is suitable for measuring the strength of the linear relationship
between two continuous variables (Khamis, n.d.). Pearson Correlation Coefficient was applied
in data analysis to explore the correlation strength between all the factors. This study also used
stepwise linear regression and random forest to identify the relationship between bio-signals
and IAQ parameters (CO2 and PM2.5) and determine the more related bio-signal. Moreover,
two ways were applied in this study for further verifying the prediction accuracy of the model
for two survey data analysis results. The first way was cross validation, we randomly chose 80%
of the data from each participant’s dataset for training the model and then used other 20% of
data to test the accuracy of the model. The second way is choosing first day’s data from each
participants’ dataset to train the model and chose the last day’s data to test the accuracy of the
model.
We also analyzed the survey data through Minitab and python to determine the most relevant
bio-signal (among heart rate, stress level, EDA and skin temperature) to IAQ evaluation and
thermal comfort, the most relevant factor (from four bio-signals, indoor temperature, indoor
relative humidity, CO2, and PM 2.5) associated with IAQ Evaluation and the factor (from four
bio-signals, indoor temperature, indoor relative humidity) most associated to thermal comfort
for each subject.
4.1 Pearson Correlation Coefficient between all parameters
Pearson correlation coefficient (rXY) is defined as a statistical measure of the degree of linear
correlation between two variables X and Y (X is the independent variable, Y is the dependent
variable) (Profillidis & Botzoris, 2019), and its analytical value is:

The range of Pearson's correlation coefficient is [-1, +1]. The value rXY = +1 indicates a perfect
positive correlation between variables X and Y, while when rXY = 0, it reflects that no
correlation between X and Y can be found (based on data and observation results). The value
rXY is equal to -1 indicates a perfect negative correlation between X and Y (Profillidis &
Botzoris, 2019).
The value range of rXY can be divided into the following other cases: 0.8< rXY <1 represents a
strong positive correlation between Y and X; 0.3< rXY <0.6 represents a moderate correlation

32

between Y and X; 0< rXY <0.3 indicates a weak positive correlation between Y and X; 0< rXY
<-0.3 testifies a weak negative correlation between Y and X; -0.3< rXY <-0.6 represents a
moderate negative correlation between Y and X, -0.8< rXY <-1 represents a strong negative
correlation between Y and X (Profillidis & Botzoris, 2019).
For each participant to understand the effects of the indoor environment, especially the effects
of the IAQ on human physiological response, we calculated the Pearson correlation coefficient
between their indoor environmental parameters (temperature, humidity, CO2, PM2.5, sound)
their physiological signals through python. Fig 4.1, 4.4. 4.7, 4.10, 4.13, 4,16. 4.19, 4.22, 4.25,
4.28, 4.31, and 4.33 respectively show the heat map results of the correlation between each
participants’ bio-signals and IAQ parameters; the IAQ parameters not only include indoor CO2
concentration and indoor PM2.5 concentration but also include the PM2.5 concentration
difference between indoor and outdoor. Besides, Fig 4.3, 4.6. 4.9, 4.12, 4.15, 4,18. 4.21, 4.24,
4.27, 4.30, 4.33, and 4.36 respectively show the heat map results of the correlation between
each participants’ bio-signals and other IEQ parameters (temperature, relative humidity,
acoustic) and survey data (IAQ evaluation and thermal comfort). Although this study measured
indoor lighting, we did not include lighting data during the data analysis part because the sensor
measuring lighting could not record data.
4.4.1 Participant 1
The experiment period of the first participant was from November 13th, 0:00 AM to November
18th, 0:00 AM, and Table 4.1 shows the descriptive statistics table created from Minitab
software for participant 1.
Table 4. 1 Descriptive Statistics Table for Participant 1

As Fig 4.1 illustrates, all the Pearson correlation coefficients between IAQ parameters and bio-
signals are less than 0.6 or larger than -0.6, indicating that all the IAQ parameters and four bio-
signals of participant 1 have a relatively weak correlation. Besides, among the four bio-signals,

33

the Pearson correlation coefficients between all the bio-signals and PM 2.5 are higher than the
Pearson correlation coefficients between those four bio-signals and CO2, which proves that the
first subject is more sensitive to indoor PM 2.5 concentration.
Furthermore, for the concentration of CO2, the Pearson correlation coefficient of stress level
is 0.2, larger than the Pearson correlation coefficient of the other three bio-signals, so stress
level is more correlated with indoor CO2 concentration among four bio-signals. In the same
way, heart rate is more correlated to indoor PM 2.5 concentration. Scatter plots of linear
regression between IAQ parameters and four bio signals are shown in Fig 4.2.

Figure. 4 1 Heat Map of Pearson Correlation between IAQ Parameters and Bio-signals -Participant 1

34

Figure. 4 2 Scatter plots of Linear regression between IAQ Parameters and Bio-signals -Participant 1
As Fig 4.3 shows, the Pearson correlation coefficients between skin temperature and indoor
temperature is 0.6, testifying a moderate positive correlation between skin temperature of
subject 1 and indoor temperature. In addition, the correlation between bio-signals and the other
factors is relatively weak.

Figure. 4 3 Heat Map of Pearson Correlation between other IEQ Parameters and survey data and Bio-signals -Participant 1
4.4.2 Participant 2
The experiment period of participate 2 was from November 19th, 3:00 PM to November 24th,
3:00 PM, and the descriptive statistics table for participant 2 is shown in Table 4.2.

35

Table 4. 2 Descriptive Statistics Table for Participant 2

Fig 4.4 shows that the Pearson correlation coefficients between all IAQ parameters and four
bio-signals are less than 0.6 or more than -0.6, which indicates a relatively weak correlation
between all the IAQ parameters and four bio-signals of participant 2. Unlike the first participant,
the EDA of participant 2 is positively correlated with CO2 and PM2.5, and the correlation
between indoor CO2 concentration and skin temperature is negative.
Furthermore, for the concentration of CO2, the Pearson correlation coefficient of EDA is 0.4,
larger than the Pearson correlation coefficient of the other three bio-signals, meaning EDA is
more correlated with indoor CO2 concentration. In the same way, EDA is also more related to
indoor PM 2.5 concentration. Scatter plots of linear regression between IAQ parameters and
four bio signals are shown in Fig 4.5.

Figure. 4 4 Heat Map of Correlation between IAQ Parameters and Bio-signals -Participant 2

36

Figure. 4 5 Scatter plots of Linear regression between IAQ Parameters and Bio-signals -Participant 2
According to Fig 4.6, none of the Pearson correlation coefficients between the four bio-signals
of participant 2 and other IEQ parameters and survey data is larger than 0.6 or less than -0.6,
which shows a relative week correlation between these parameters.

Figure. 4 6 Heat Map of Correlation between other IEQ Parameters and survey data and Bio-signals -Participant 2

37

4.4.3 Participant 3
The experiment period of participate 3 was from November 19th, 3:00 PM to November 24th,
3:00 PM. Besides, the second and third subjects lived in the same room, and they did this
experiment in the same five days. Table 4.3 shows the descriptive statistics table created from
Minitab software for participant 3.
Table 4. 3 Descriptive Statistics Table for Participant 3

As Fig 4.7 illustrates, the Pearson correlation coefficients between PM2.5 and skin temperature
are -0.6, indicating a moderate negative correlation between them. Except for the correlation
between skin temperature and PM2.5, the correlation between IAQ parameters and four bio-
signals of participant 3 is relatively weak. Besides, the Pearson correlation coefficients between
three of the four bio-signals (heart rate, stress level, and skin temperature) and PM 2.5 are
higher than the Pearson correlation coefficients between these three bio-signals and CO2.
Furthermore, for the concentration of CO2, the Pearson correlation coefficient of skin
temperature is -0.5, larger than the Pearson correlation coefficient of the other three bio-signals,
so skin temperature is more correlated with indoor CO2 concentration among four bio-signals.
In the same way, skin temperature is also more correlated to indoor PM 2.5 concentration.
Scatter plots of linear regression between IAQ parameters and four bio signals are shown in
Fig 4.8.

38

Figure. 4 7 Heat Map of Correlation between IAQ Parameters and Bio-signals -Participant 3

Figure. 4 8 Scatter plots of Linear regression between IAQ Parameters and Bio-signals -Participant 3
Fig 4.9 is the heat map of correlation between bio-signals and other IEQ parameters of
participant 3, the correlation between all the bio signals and other IEQ factors is relatively weak
because none of the Pearson correlation coefficients is larger than 0.6 or less than -0.6.

39

Figure. 4 9 Heat Map of Correlation between other IEQ Parameters and Bio-signals -Participant 3
The analysis process of the other nine participants is the same as that of the first three
participants.
4.2 Investigate the most relevant bio-signal to IAQ parameters
Stepwise Linear Regression and Random Forest were used in all the analyses in this part. While
exploring the more relevant factors to IAQ Evaluation and thermal comfort, if more than four
relevant factors appeared on the stepwise linear regression result table created from Minitab,
we chose the first four parameters as the more relevant factors to IAQ Evaluation and thermal
comfort. Although bio-signals cannot cause the variation of IAQ parameters, we chose four
bio-signals as independent variable and IAQ parameters as dependent variable in stepwise
regression analysis process. Because in this chapter, we mainly explored the correlation degree
between four bio-signals and IAQ parameters, for finding bio-signals more relevant to IAQ
parameters.
4.2.1 Participant 1
According to the order of step distribution shown in Fig 4.37 (208 rows of the dataset of
participant 1 were used), stress level is most significantly related to the CO2 concentration, and
skin temperature and heart rate are non-significant factors to CO2 concentration because they
did not appear on the stepwise analysis result table. Besides, based on the R-sq value, R-sq for
stress level is 2.84%, which means 2.84% of fluctuation of CO2 concentration is associated
with stress level.

40

Figure. 4 10 Stepwise Linear Regression Analysis Result (Dependent variable: CO2, Independent variables: four bio-
signals) -Participant 1
Fig 4.38 shows the stepwise linear regression analysis results for identifying the relationship
between PM2.5 and four bio-signals (208 rows of the dataset of participant 1 were used); based
on the order of step distribution, heart rate is most significantly related to the indoor PM2.5
concentration. Furthermore, the P-value of heart rate proves a statistically significant
relationship between skin temperature and PM 2.5 concentration in the bedroom of participant
1. Besides, based on the R-sq value, R-sq for skin temperature is 20.62%, which means 20.62%
of the variation of PM 2.5 concentration is related to skin temperature.

Figure. 4 11 Stepwise Linear Regression Analysis Result (Dependent variable: PM2.5, Independent variables: four bio-
signals) -Participant 1
4.2.2 Participant 2
According to Fig 4.39 (422 rows of the dataset of participant 2 were used), the P-value of stress
level, heart rate, and EDA are 0, smaller than 0.5, which means that the relationship between
CO2 concentration and these three bio-signals are significantly related, but stress level is the
most important factor. Because the skin temperature does not appear on the analysis result
sheet, it is a non-significant factor to CO2 concentration.

41

Figure. 4 12 Stepwise Linear Regression Analysis Result (Dependent variable: CO2, Independent variables: four bio-
signals) -Participant 1
According to Fig 4.40 (337 rows of the dataset of participant 2 were used), EDA contributes
most to the PM2.5 concentration. Besides, the P-value of EDA is 0, which testifies that the
relationship between EDA of participant 2 and PM 2.5 concentration is statistically significant.
Also, R-sq for skin temperature is 22.6%, which means 22.6% of the variation of PM2.5
concentration is related to EDA. Besides, stress level and skin temperature have no significant
relationship with PM 2.5 concentration because they are not shown in the result table.

Figure. 4 13 Stepwise Linear Regression Analysis Result (Dependent variable: PM2.5, Independent variables: four bio-
signals) -Participant 2
4.2.3 Participant 3
According to the order of step distribution shown in Fig 4.41 (258 rows of the dataset of
participant 3 were used), skin temperature is the most important bio-signal (among four bio-
signals) to the CO2 concentration, and stress level is a non-significant factor. Also, based on
the R-sq value of skin temperature, 26.96% of fluctuation of CO2 concentration is associated
with skin temperature.

42

Figure. 4 14 Stepwise Linear Regression Analysis Result (Dependent variable: CO2, Independent variables: four bio-
signals) -Participant 3
Fig 4.42 shows the stepwise linear regression analysis results for identifying the relationship
between PM2.5 and four bio-signals (202 rows of the dataset of participant 3 were used).
According to this analysis result, skin temperature is the most relevant bio-signal to indoor
PM2.5 concentration, and stress level is least significantly contributed to the PM 2.5
concentration. Also, the relationship between skin temperature of participant 1 and PM 2.5
concentration is statistically significant because the P-value of skin temperature is 0. In addition,
R-sq for skin temperature is 35.92%, indicating that 35.92% of the variation of PM 2.5
concentration is related to skin temperature.

Figure. 4 15 Stepwise Linear Regression Analysis Result (Dependent variable: PM2.5, Independent variables: four bio-
signals) -Participant 3
The analysis process of the other nine participants is the same as that of the first three
participants.

43

4.3 Investigate the most relevant bio-signal to IAQ Evaluation
4.3.1 Participant 1
Based on stepwise linear regression analysis result showing in Fig 16 (140 rows of the dataset
of participant 1 were used), among four bio-signals, stress level and skin temperature are the
most relevant bio-signals to the IAQ Evaluation, but stress level is the most relevant one.

Figure. 4 16 Stepwise Linear Regression Analysis Result (Dependent variable: IAQ Evaluation, Independent variables:
heart rate, stress level, skin temperature, EDA) -Participant 1
According to the evaluation of the random forest model of using stress level and skin
temperature to predict IAQ Evaluation of participant 1, the accuracy of the model is around 90%
based on cross validation (randomly chose 80% data from dataset to train the model, and used
the other 20% data from the dataset to test the model) is shown in Fig 17.

Figure. 4 17 Random Forest Algorithm and Accuracy of the model (Use stress level and skin temperature to predict IAQ
Evaluation, use cross validation to test accuracy)-Participant 1
Furthermore, we also evaluate the accuracy of the random forest model of using all bio-signals
to predict IAQ Evaluation of participant 1. The accuracy is around 89% (Fig 4.18), lower than
90%.

44

Figure. 4 18 Random Forest Algorithm and Accuracy of the model (Use all bio-signals to predict IAQ Evaluation, use cross
validation to test accuracy)-Participant 1
For the second way to test accuracy of the model of using stress level and skin temperature to
predict IAQ Evaluation, we chose first day’s data to train a model, and chose the last day’s data
to test the model. The accuracy is 93% (Fig 4.19), higher than 90% (accuracy based on cross
validation).

Figure. 4 19 Random Forest Algorithm for using first day to train a model and last day data to test accuracy of the model
(Use stress level and skin temperature to predict IAQ Evaluation)-Participant 1
In the same way, the accuracy of the random forest model of using all bio-signals to predict
IAQ Evaluation of participant 1 is around 97% (Fig 4.20), higher than 89% (accuracy based on
cross validation), and higher than 93% (use more relevant bio-signals to predict IAQ
Evaluation).

Figure. 4 20 Random Forest Algorithm for using first day to train a model and last day data to test accuracy of the model
(Use all bio-signals to predict IAQ Evaluation)-Participant 1
We also used random forest to investigate the importance of four bio-signals affecting IAQ
Evaluation, the result is shown in Fig 4.21. Among four bio-signals, heart rate and stress level

45

are the more relevant bio-signals to the IAQ Evaluation, but heart rate is the most relevant one.

Figure. 4 21 Feature importance (four bio-signals) graph from Random Forest - Participant 1
The accuracy of the model of using heart rate and stress level is around 83% based on cross
validation (Fig 4.22), lower than 90% (accuracy based on stepwise linear regression analysis
result), and lower than 89% (accuracy of the model of using all bio-signals to predict IAQ
Evaluation).

Figure. 4 22 Random Forest Algorithm and Accuracy of the model (Use heart rate and stress level to predict IAQ
Evaluation, use cross validation to test accuracy)-Participant 1
For the second way to test accuracy of the model of using stress level and skin temperature to
predict IAQ Evaluation (chose first day’s data to train a model and chose the last day’s data to
test the model). The accuracy is around 92% (Fig 4.23), higher than 90% (accuracy based on
cross validation), and lower than 97% (accuracy of the model of using all bio-signals to predict
IAQ Evaluation).

46

Figure. 4 23 Random Forest Algorithm for using first day to train a model and last day data to test accuracy of the model
(Use stress level and skin temperature to predict IAQ Evaluation)-Participant 1
Table 4.4 shows the summary of the more relevant bio-signals based on stepwise linear
regression analysis result and random forest result.
Table 4. 4 Summary of relevant bio-signals based on two different methods - Participant 1
Method More relevant bio-signals
Stepwise linear regression Stress level, skin temperature
Random forest Heart rate, stress level
Table 4.5 shows the summary of accuracy of different IAQ_Evaluation prediction model, in the table, S
means the accuracy based on the stepwise linear regression result, R means the accuracy based on the random
forest result.
Table 4. 5 Summary of accuracy of different IAQ_Evaluation prediction model - Participant 1
Model details and way of testing accuracy Accuracy of the model
S - Use relevant bio-signals to predict (cross validation) 90%
S - Use relevant bio-signals to predict (first day and last day) 93%
R - Use relevant bio-signals to predict (cross validation) 83%
R - Use relevant bio-signals to predict (first day and last day) 92%
Use all bio-signals to predict (cross validation) 89%
Use all bio-signals to predict (first day and last day) 97%
4.3.2 Participant 2
Fig 4.24 shows the stepwise linear regression analysis result (420 rows of the dataset of
participant 2 were used), illustrating that among four bio-signals, only heart rate is significantly
associated with IAQ Evaluation of participant 2.

47

Figure. 4 24 Stepwise Linear Regression Analysis Result (Dependent variable: IAQ Evaluation, Independent variables:
heart rate, EDA, skin temperature, and stress level) -Participant 2
The accuracy of the model of using heart rate to predict AQ Evaluation of participant 2 is
around 96.6% based on cross validation (randomly chose 80% data from dataset to train the
model, and used the other 20% data from the dataset to test the model), which is shown in Fig
4.25.

Figure. 4 25 Random Forest Algorithm and Accuracy of the model (Use heart rate to predict IAQ Evaluation)-Participant 2
The accuracy of the random forest model of using all bio-signals to predict IAQ Evaluation of
participant 2 is around 96.4% (Fig 4.26), lower than 96.6%.

Figure. 4 26 Random Forest Algorithm and Accuracy of the model (Use all bio-signals to predict IAQ Evaluation)-
Participant 2
For the second way to test accuracy of the model of using heart rate to predict IAQ Evaluation

48

(chose first day’s data to train a model, and chose the last day’s data to test the model). The
accuracy is 96.6% (Fig 4.27), equal to 96.6% (accuracy based on cross validation).

Figure. 4 27 Random Forest Algorithm for using first day to train a model and last day data to test accuracy of the model
(Use heart rate to predict IAQ Evaluation)-Participant 2
In the same way, the accuracy of the random forest model of using all bio-signals to predict
IAQ Evaluation of participant 2 is around 96.7% (Fig 4.28), higher than 96.6% (accuracy based
on cross validation), and higher than 96.6% (use more relevant bio-signals to predict IAQ
Evaluation-accuracy based on first day and last day data).

Figure. 4 28 Random Forest Algorithm for using first day to train a model and last day data to test accuracy of the model
(Use all bio-signals to predict IAQ Evaluation)-Participant 2
We also used random forest to investigate the importance of four bio-signals affecting IAQ
Evaluation, the result is shown in Fig 4.29. Among four bio-signals, EDA is more relevant bio-
signal to the IAQ Evaluation.

Figure. 4 29 Feature importance (four bio-signals) graph from Random Forest - Participant 2

49

The accuracy of the model of using EDA is around 96.7% based on cross validation (Fig 4.30),
higher than 96.6% (accuracy based on stepwise linear regression analysis result), and higher
than 96.4% (accuracy of the model of using all bio-signals to predict IAQ Evaluation).

Figure. 4 30 Random Forest Algorithm and Accuracy of the model (Use EDA to predict IAQ Evaluation, use cross
validation to test accuracy)-Participant 2
For the second way to test accuracy of the model of using EDA to predict IAQ Evaluation
(chose first day’s data to train a model and chose the last day’s data to test the model). The
accuracy is around 96.2% (Fig 4.31), lower than 96.7% (accuracy based on cross validation),
and lower than 96.7% (accuracy of the model of using all bio-signals to predict IAQ
Evaluation).

Figure. 4 31 Random Forest Algorithm for using first day to train a model and last day data to test accuracy of the model
(Use EDA to predict IAQ Evaluation)-Participant 2
Table 4.6 shows the summary of the more relevant bio-signals based on stepwise linear
regression analysis result and random forest result.
Table 4. 6 Summary of relevant bio-signals based on two different methods - Participant 2
Method More relevant bio-signals
Stepwise linear regression Heart rate
Random forest EDA
Table 4.7 shows the summary of accuracy of different IAQ_Evaluation prediction model, in
the table, S means the accuracy based on the stepwise linear regression result, R means the
accuracy based on the random forest result.
Table 4. 7 Summary of accuracy of different IAQ_Evaluation prediction model - Participant 2

50

Model details and way of testing accuracy Accuracy of the model
S - Use relevant bio-signals to predict (cross validation) 96.6%
S - Use relevant bio-signals to predict (first day and last day) 96.6%
R - Use relevant bio-signals to predict (cross validation) 96.7%
R - Use relevant bio-signals to predict (first day and last day) 96.2%
Use all bio-signals to predict (cross validation) 96.4%
Use all bio-signals to predict (first day and last day) 96.7%
4.3.3 Participant 3
As Fig 4.32 illustrated (181 rows of the dataset of participant 3 were used), all four bio-signals
are significantly associated with IAQ Evaluation of participant 3, but stress level is the most
relevant bio-signal based on the stepwise linear regression result.

Figure. 4 32 Stepwise Linear Regression Analysis Result (Dependent variable: IAQ Evaluation, Independent variables:
heart rate, stress level, skin temperature, EDA) -Participant 3
The accuracy of the model of using four bio-signals to predict IAQ Evaluation of participant 3
is around 84% based on cross validation (randomly chose 80% data from dataset to train the
model, and used the other 20% data from the dataset to test the model), which is shown in Fig
4.33.

Figure. 4 33 Random Forest Algorithm and Accuracy of the model (Use all bio-signals to predict IAQ Evaluation)-
Participant 3
For the second way to test accuracy of the model of using heart rate to predict IAQ Evaluation

51

(chose first day’s data to train a model, and chose the last day’s data to test the model). The
accuracy is 91% (Fig 4.34), equal to 96.6% (accuracy based on cross validation).

Figure. 4 34 Random Forest Algorithm for using first day to train a model and last day data to test accuracy of the model
(Use heart rate to predict IAQ Evaluation)-Participant 2
We also used random forest to investigate the importance of four bio-signals affecting IAQ
Evaluation, the result is shown in Fig 4.35. Among four bio-signals, stress level and skin
temperature are more relevant bio-signals to the IAQ Evaluation.

Figure. 4 35 Feature importance (four bio-signals) graph from Random Forest - Participant 3
The accuracy of the model of using EDA and stress level is around 81% based on cross
validation (Fig 4.36), lower than 84% (accuracy based on stepwise linear regression analysis
result).

Figure. 4 36 Random Forest Algorithm and Accuracy of the model (Use EDA and stress level to predict IAQ Evaluation, use
cross validation to test accuracy)-Participant 3

52

For the second way to test accuracy of the model of using EDA and stress level to predict IAQ
Evaluation (chose first day’s data to train a model and chose the last day’s data to test the
model). The accuracy is around 86% (Fig 4.37), higher than 81% (accuracy based on cross
validation), and lower than 91% (accuracy of the model of using all bio-signals to predict IAQ
Evaluation).

Figure. 4 37 Random Forest Algorithm for using first day to train a model and last day data to test accuracy of the model
(Use EDA to predict IAQ Evaluation)-Participant 3
Table 4.8 shows the summary of the more relevant bio-signals based on stepwise linear
regression analysis result and random forest result.
Table 4. 8 Summary of relevant bio-signals based on two different methods - Participant 3
Method More relevant bio-signals
Stepwise linear regression Four bio-signals
Random forest stress level, skin temperature
Table 4.9 shows the summary of accuracy of different IAQ_Evaluation prediction model, in
the table, S means the accuracy based on the stepwise linear regression result, R means the
accuracy based on the random forest result.
Table 4. 9 Summary of accuracy of different IAQ_Evaluation prediction model - Participant 3
Model details and way of testing accuracy Accuracy of the model
S - Use relevant bio-signals to predict (cross validation) 84%
S - Use relevant bio-signals to predict (first day and last day) 91%
R - Use relevant bio-signals to predict (cross validation) 81%
R - Use relevant bio-signals to predict (first day and last day) 86%
Use all bio-signals to predict (cross validation) 84%
Use all bio-signals to predict (first day and last day) 91%
The analysis process of the other nine participants is the same as that of the first three
participants.

53

4.4 Investigate the most relevant factors (including bio-signals, indoor temp, indoor RH and
IAQ parameters) to IAQ Evaluation
4.4.1 Participant 1
According to the order of step showing in the Fig 4.38 (141 rows of the dataset of participant
1 were used), among eight independent variables (four bio-signals, indoor temperature, indoor
relative humidity, CO2, and PM 2.5), four factors, PM2.5, stress level, CO2 and indoor
temperature are relatively associated with IAQ evaluation of participant 1, and PM2.5 is the
most relevant factor.

Figure. 4 38 Stepwise Linear Regression Analysis Result (Dependent variable: IAQ Evaluation, Independent variables: four
bio-signals, indoor temperature, indoor relative humidity, indoor CO2, and indoor PM 2.5) -Participant 1
According to the evaluation of the random forest model of using PM2.5, stress level, CO2, and
indoor temperature to predict IAQ Evaluation of participant 1, the accuracy of the model is
around 91% based on cross validation (randomly chose 80% data from dataset to train the
model, and used the other 20% data from the dataset to test the model) is shown in Fig 4.39.

Figure. 4 39 Random Forest Algorithm and Accuracy of the model (Use indoor PM2.5, stress level, indoor CO2, and indoor
temperature to predict IAQ Evaluation)-Participant 1
The accuracy of random forest model of using all eight parameters to predict IAQ Evaluation

54

of participant 1 is around 89% (Fig 4.40), lower than 91%.

Figure. 4 40 Random Forest Algorithm and Accuracy of the model (Use all eight parameters to predict IAQ Evaluation)-
Participant 1
For the second way to test accuracy of the model of using PM2.5, stress level, CO 2, and indoor
temperature to predict IAQ Evaluation, we chose first day’s data to train a model, and chose
the last day’s data to test the model. The accuracy is 97% (Fig 41), higher than 91% (accuracy
based on cross validation).

Figure. 4 41 Random Forest Algorithm for using first day to train a model and last day data to test accuracy of the model
(Use PM2.5, stress level, CO2, and indoor temperature to predict IAQ Evaluation)-Participant 1
In the same way, the accuracy of the random forest model of using all eight parameters to
predict IAQ Evaluation of participant 1 is around 99% (Fig 4.42), higher than 89% (accuracy
based on cross validation), and higher than 97% (use more relevant parameters to predict IAQ
Evaluation-accuracy based on first day and last day data).

Figure. 4 42 Random Forest Algorithm for using first day to train a model and last day data to test accuracy of the model
(Use all eight parameters to predict IAQ Evaluation)-Participant 1
We also used random forest to investigate the importance of all eight parameters affecting IAQ
Evaluation, the result is shown in Fig 4.43. Among eight parameters, PM2.5, stress level, CO 2

55

and heart rate are more relevant parameters to the IAQ Evaluation.

Figure. 4 43 Feature importance (eight parameters) graph from Random Forest - Participant 1
The accuracy of the model of using PM2.5, stress level, CO 2 and heart rate to predict IAQ
Evaluation is around 86% based on cross validation (Fig 4.44), lower than 91% (accuracy based
on stepwise linear regression analysis result), and lower than 89% (accuracy of the model of
using all eight parameters to predict IAQ Evaluation – cross validation).

Figure. 4 44 Random Forest Algorithm and Accuracy of the model (Use PM2.5, stress level, CO2 and heart rate to predict
IAQ Evaluation, use cross validation to test accuracy)-Participant 1
For the second way to test accuracy of the model of using PM2.5, stress level, CO 2 and heart
rate to predict IAQ Evaluation (chose first day’s data to train a model and chose the last day’s
data to test the model). The accuracy is around 98% (Fig 4.45), higher than 86% (accuracy
based on cross validation), and lower than 99% (accuracy of the model of using all parameters
to predict IAQ Evaluation).

56

Figure. 4 45 Random Forest Algorithm for using first day to train a model and last day data to test accuracy of the model
(Use PM2.5, stress level, CO2 and heart rate to predict IAQ Evaluation)-Participant 1
Table 4.10 shows the summary of the more relevant parameters based on stepwise linear
regression analysis result and random forest result.
Table 4.10 Summary of relevant parameters based on two different methods - Participant 1
Method More relevant parameters
Stepwise linear regression PM2.5, stress level, CO2, and indoor temperature
Random forest PM2.5, stress level, CO2, and heart rate
Table 4.11 shows the summary of accuracy of different IAQ_Evaluation prediction model, in
the table, S means the accuracy based on the stepwise linear regression result, R means the
accuracy based on the random forest result.
Table 4.11 Summary of accuracy of different IAQ_Evaluation prediction model - Participant 1
Model details and way of testing accuracy Accuracy of the model
S - Use relevant parameters to predict (cross validation) 91%
S - Use relevant parameters to predict (first day and last day) 96%
R - Use relevant parameters to predict (cross validation) 86%
R - Use relevant parameters to predict (first day and last day) 98%
Use all parameters to predict (cross validation) 89%
Use all parameters to predict (first day and last day) 99%
4.4.2 Participant 2
Fig 4.46 (335 rows of the dataset of participant 2 were used) shows the stepwise linear
regression analysis result for identifying the relationship between IAQ Evaluation and eight
parameters (four bio-signals, indoor temperature, indoor relative humidity, CO 2, and PM 2.5).
According to the order of step, four parameters, indoor temperature, PM2.5, indoor relative
humidity, and EDA are more associated with IAQ evaluation of participant 2, and indoor
temperature is the most relevant factor.

57

Figure. 4 46 Stepwise Linear Regression Analysis Result (Dependent variable: IAQ Evaluation, Independent variables: four
bio-signals, temperature, relative humidity, CO2, and PM 2.5) -Participant 2
Based on the evaluation of the random forest model of using indoor temperature shown in Fig
4.47, PM2.5, indoor relative humidity and EDA to predict IAQ Evaluation of participant 2, the
accuracy of the model is around 98% based on cross validation (randomly chose 80% data from
dataset to train the model, and used the other 20% data from the dataset to test the model).

Figure. 4 47 Random Forest Algorithm and Accuracy of the model (Use indoor temperature, indoor PM2.5, indoor relative
humidity, and EDA to predict IAQ Evaluation)-Participant 2
The accuracy of random forest model of using all eight parameters to predict IAQ Evaluation
of participant 2 is around 99% (Fig 4.48), lower than 98%.

Figure. 4 48 Random Forest Algorithm and Accuracy of the model (Use all eight parameters to predict IAQ Evaluation)-
Participant 2

58

For the second way to test accuracy of the model of using indoor temperature, PM2.5, indoor
relative humidity and EDA to predict IAQ Evaluation, we chose first day’s data to train a model,
and chose the last day’s data to test the model. The accuracy is 99% (Fig 4.49), higher than 98%
(accuracy based on cross validation).
Figure. 4 49 Random Forest Algorithm for using first day to train a model and last day data to test accuracy of the model
(Use PM2.5, stress level, CO2, and indoor temperature to predict IAQ Evaluation)-Participant 2
In the same way, the accuracy of the random forest model of using all eight parameters to
predict IAQ Evaluation of participant 2 is around 98% (Fig 4.50), lower than 99% (accuracy
based on cross validation), and lower than 99% (use more relevant parameters to predict IAQ
Evaluation-accuracy based on first day and last day data).

Figure. 4 50 Random Forest Algorithm for using first day to train a model and last day data to test accuracy of the model
(Use all eight parameters to predict IAQ Evaluation)-Participant 2
We also used random forest to investigate the importance of all eight parameters affecting IAQ
Evaluation, the result is shown in Fig 4.51. Among eight parameters, EDA, CO2, indoor
temperature, and PM2.5, are more relevant parameters to the IAQ Evaluation.

Figure. 4 51 Feature importance (eight parameters) graph from Random Forest - Participant 2

59

The accuracy of the model of using EDA, CO2, indoor temperature, and PM2.5 to predict IAQ
Evaluation is around 99% based on cross validation (Fig 4.52), higher than 98% (accuracy
based on stepwise linear regression analysis result), and lower than 99% (accuracy of the model
of using all eight parameters to predict IAQ Evaluation – cross validation).

Figure. 4 52 Random Forest Algorithm and Accuracy of the model (Use EDA, CO2, indoor temperature, and PM2.5 to
predict IAQ Evaluation, use cross validation to test accuracy)-Participant 2
For the second way to test accuracy of the model of using EDA, CO2, indoor temperature, and
PM2.5 to predict IAQ Evaluation (chose first day’s data to train a model and chose the last
day’s data to test the model). The accuracy is around 98% (Fig 4.52), lower than 99% (accuracy
based on cross validation), and equal to 98% (accuracy of the model of using all parameters to
predict IAQ Evaluation).

Figure. 4 53 Random Forest Algorithm for using first day to train a model and last day data to test accuracy of the model
(Use EDA, CO2, indoor temperature, and PM2.5 to predict IAQ Evaluation)-Participant 2
Table 4.12 shows the summary of the more relevant parameters based on stepwise linear
regression analysis result and random forest result.
Table 4.12 Summary of relevant parameters based on two different methods - Participant 2
Method More relevant parameters
Stepwise linear regression Indoor temperature, PM2.5, indoor relative humidity and EDA
Random forest EDA, CO2, indoor temperature, and PM2.5
Table 4.13 shows the summary of accuracy of different IAQ_Evaluation prediction model, in
the table, S means the accuracy based on the stepwise linear regression result, R means the
accuracy based on the random forest result.

60

Table 4.13 Summary of accuracy of different IAQ_Evaluation prediction model - Participant 2
Model details and way of testing accuracy Accuracy of the model
S - Use relevant parameters to predict (cross validation) 98%
S - Use relevant parameters to predict (first day and last day) 99%
R - Use relevant parameters to predict (cross validation) 99%
R - Use relevant parameters to predict (first day and last day) 98%
Use all parameters to predict (cross validation) 99%
Use all parameters to predict (first day and last day) 98%
4.4.3 Participant 3
As Fig 4.54 illustrated, four parameters, indoor relative humidity, stress level, skin temperature,
and indoor temperature are more associated with IAQ evaluation of participant 3, and indoor
relative humidity is the most relevant factor.

Figure. 4 54 Stepwise Linear Regression Analysis Result (Dependent variable: IAQ Evaluation, Independent variables: four
bio-signals, temperature, relative humidity, CO2, and PM 2.5) -Participant 3
According to the evaluation of the random forest model of using indoor relative humidity, stress
level, skin temperature, and indoor temperature to predict IAQ Evaluation of participant 3, the
accuracy of the model is around 90% based on cross validation (randomly chose 80% data from
dataset to train the model, and used the other 20% data from the dataset to test the model) is
shown in Fig 4.55.

61

Figure. 4 55 Random Forest Algorithm and Accuracy of the model (Use indoor relative humidity, stress level, skin
temperature, and indoor temperature to predict IAQ Evaluation)-Participant 3
The accuracy of random forest model of using all eight parameters to predict IAQ Evaluation
of participant 3 is around 94% (Fig 4.56), higher than 90%.

Figure. 4 56 Random Forest Algorithm and Accuracy of the model (Use all eight parameters to predict IAQ Evaluation)-
Participant 3
For the second way to test accuracy of the model of using indoor relative humidity, stress level,
skin temperature, and indoor temperature to predict IAQ Evaluation, we chose first day’s data
to train a model, and chose the last day’s data to test the model. The accuracy is 95% (Fig 4.57),
higher than 90% (accuracy based on cross validation).

Figure. 4 57 Random Forest Algorithm for using first day to train a model and last day data to test accuracy of the model
(Use indoor relative humidity, stress level, skin temperature, and indoor temperature to predict IAQ Evaluation) -
Participant 3
In the same way, the accuracy of the random forest model of using all eight parameters to

62

predict IAQ Evaluation of participant 3 is around 99% (Fig 4.58), higher than 94% (accuracy
based on cross validation), and higher than 95% (use more relevant parameters to predict IAQ
Evaluation-accuracy based on first day and last day data).

Figure. 4 58 Random Forest Algorithm for using first day to train a model and last day data to test accuracy of the model
(Use all eight parameters to predict IAQ Evaluation)-Participant 3
We also used random forest to investigate the importance of all eight parameters affecting IAQ
Evaluation, the result is shown in Fig 4.59. Among eight parameters, indoor relative humidity,
PM2.5, CO2, and skin temperature are more relevant parameters to the IAQ Evaluation.

Figure. 4 59 Feature importance (eight parameters) graph from Random Forest - Participant 3
The accuracy of the model of using indoor relative humidity, PM2.5, CO 2, and skin temperature
to predict IAQ Evaluation is around 98% based on cross validation (Fig 4.60), higher than 90%
(accuracy based on stepwise linear regression analysis result), and higher than 94% (accuracy
of the model of using all eight parameters to predict IAQ Evaluation – cross validation).

63

Figure. 4 60 Random Forest Algorithm and Accuracy of the model (Use indoor relative humidity, PM2.5, CO2, and skin
temperature to predict IAQ Evaluation, use cross validation to test accuracy)-Participant 3
For the second way to test accuracy of the model of using indoor relative humidity, PM2.5,
CO2, and skin temperature to predict IAQ Evaluation (chose first day’s data to train a model
and chose the last day’s data to test the model). The accuracy is around 99.6% (Fig 4.4.24),
higher than 98% (accuracy based on cross validation), and higher than 99% (accuracy of the
model of using all parameters to predict IAQ Evaluation).

Figure. 4 61 Random Forest Algorithm for using first day to train a model and last day data to test accuracy of the model
(Use EDA, CO2, indoor temperature, and PM2.5 to predict IAQ Evaluation)-Participant 3
Table 4.13 shows the summary of the more relevant parameters based on stepwise linear
regression analysis result and random forest result.
Table 4.13 Summary of relevant parameters based on two different methods - Participant 3
Method More relevant parameters
Stepwise linear regression Indoor relative humidity, stress level, skin temperature, and indoor temperature
Random forest Indoor relative humidity, PM2.5, CO2, and skin temperature
Table 4.14 shows the summary of accuracy of different IAQ_Evaluation prediction model, in
the table, S means the accuracy based on the stepwise linear regression result, R means the
accuracy based on the random forest result.

64

Table 4. 14 Summary of accuracy of different IAQ_Evaluation prediction model - Participant 3
Model details and way of testing accuracy Accuracy of the model
S - Use relevant parameters to predict (cross validation) 90%
S - Use relevant parameters to predict (first day and last day) 95%
R - Use relevant parameters to predict (cross validation) 98%
R - Use relevant parameters to predict (first day and last day) 99%
Use all parameters to predict (cross validation) 94%
Use all parameters to predict (first day and last day) 99%
The analysis process of the other nine participants is the same as that of the first three
participants.
4.5 Investigate the most relevant factors (including bio-signals, indoor temp and indoor RH) to
thermal comfort
4.5.1 Participant 1
As Fig 4.5.1 (131 rows of the dataset of participant 1 were used) illustrated, among six
independent variables (four bio-signals, indoor temperature, indoor relative humidity), indoor
temperature, stress level, EDA, and skin temperature are more associated with thermal comfort
of participant 1, and indoor temperature is the most relevant factor.

Figure. 4 62 Stepwise Linear Regression Analysis Result (Dependent variable: Thermal Comfort, Independent variables:
four bio-signals, temperature, relative humidity) -Participant 1
Evaluation of random forest model of using indoor temperature, stress level, EDA, and skin
temperature to predict the thermal comfort is showing in Fig 4.63, and the accuracy is around
88% based on cross validation (randomly chose 80% data from dataset to train the model, and
used the other 20% data from the dataset to test the model).

65

Figure. 4 63 Random Forest Algorithm and Accuracy of the model (Use indoor temperature, stress level, EDA, and skin
temperature to predict thermal comfort, use cross validation test accuracy)-Participant 1
The accuracy of random forest model of using all six parameters to predict thermal comfort of
participant 1 is around 73% (Fig 4.64), lower than 90%.

Figure. 4 64 Random Forest Algorithm and Accuracy of the model (Use all six parameters to predict thermal comfort, use
cross validation test accuracy)-Participant 1
For the second way to test accuracy of the model of using indoor temperature, stress level,
EDA, and skin temperature to predict thermal comfort, we chose first day’s data to train a
model, and chose the last day’s data to test the model. The accuracy is 92% (Fig 4.65), higher
than 88% (accuracy based on cross validation).

Fig 4.65 Random Forest Algorithm for using first day to train a model and last day data to test accuracy of the model (Use
indoor temperature, stress level, EDA and skin temperature to predict thermal comfort) - Participant 1

66

In the same way, the accuracy of the random forest model of using all six parameters to predict
thermal comfort of participant 1 is around 95% (Fig 4.5.5), higher than 73% (accuracy based
on cross validation), and higher than 92% (use more relevant parameters to predict thermal
comfort-accuracy based on first day and last day data).

Figure. 4 65 Random Forest Algorithm for using first day to train a model and last day data to test accuracy of the model
(Use all six parameters to predict thermal comfort)-Participant 1
We applied random forest to investigate the importance of all six parameters affecting thermal
comfort, as Fig 4.66 illustrated. Among six parameters, indoor temperature, skin temperature,
indoor relative humidity and EDA are more important parameters to thermal comfort.

Figure. 4 66 Feature importance (six parameters) graph from Random Forest - Participant 1
The accuracy of the model of using indoor temperature, skin temperature, indoor relative
humidity and EDA to predict thermal comfort is around 80% based on cross validation (Fig
4.67), lower than 88% (accuracy based on stepwise linear regression analysis result – cross
validation), and higher than 73% (accuracy of the model of using all six parameters to predict
thermal comfort – cross validation).

67

Figure. 4 67 Random Forest Algorithm and Accuracy of the model (Use indoor temperature, skin temperature, indoor
relative humidity and EDA to predict thermal comfort, use cross validation to test accuracy)-Participant 1
As for using the second way to test accuracy of the model of using indoor temperature, skin
temperature, indoor relative humidity, and EDA to predict thermal comfort (chose first day’s
data to train a model and chose the last day’s data to test the model). The accuracy is around
84% (Fig 4.68), higher than 80% (accuracy based on cross validation), and lower than 95%
(accuracy of the model of using all parameters to predict thermal comfort).

Figure. 4 68 Random Forest Algorithm for using first day to train a model and last day data to test accuracy of the model
(Use indoor temperature, skin temperature, indoor relative humidity and EDA to predict thermal comfort)-Participant 1
Table 4.15 shows the summary of the more important parameters based on stepwise linear
regression analysis result and random forest result.
Table 4. 15 Summary of relevant parameters to thermal comfort - Participant 1
(S – based on stepwise linear regression result, R – based on random forest result)
Method More relevant parameters
Stepwise linear regression Indoor relative humidity, stress level, skin temperature, and indoor temperature
Random forest Indoor temperature, skin temperature, indoor relative humidity, EDA
Table 4.16 shows the summary of accuracy of different thermal comfort prediction model, in
the table, S means the accuracy based on the stepwise linear regression result, R means the
accuracy based on the random forest result.

68

Table 4. 16 Summary of accuracy of different thermal comfort prediction model - Participant 1
(S – based on stepwise linear regression result, R – based on random forest result)
Model details and way of testing accuracy Accuracy of the model
S - Use relevant parameters to predict (cross validation) 88%
S - Use relevant parameters to predict (first day and last day) 92%
R - Use relevant parameters to predict (cross validation) 80%
R - Use relevant parameters to predict (first day and last day) 84%
Use all parameters to predict (cross validation) 73%
Use all parameters to predict (first day and last day) 95%
4.5.2 Participant 2
Based on the order of step showing in Fig 4.69 (420 rows of the dataset of participant 2 were
used), among six independent variables (four bio-signals, indoor temperature, indoor relative
humidity), skin temperature, indoor temperature, EDA, and stress level are more related with
thermal comfort of participant 2, and skin temperature is the most relevant factor.

Figure. 4 69 Stepwise Linear Regression Analysis Result (Dependent variable: Thermal Comfort, Independent variables:
four bio-signals, temperature, relative humidity) -Participant 2
Fig 4.70 shows the random forest model for using skin temperature, indoor temperature, EDA
and stress level to predict the thermal comfort (based on stepwise results), and the accuracy
of the model is around 82% based on cross validation (randomly chose 80% data from dataset
to train the model, and used the other 20% data from the dataset to test the model).

69

Figure. 4 70 Random Forest Algorithm and Accuracy of the model (Use skin temperature, indoor temperature, EDA, and
stress level to predict thermal comfort)-Participant 2
The accuracy of random forest model of using all six parameters to predict thermal comfort of
participant 2 is around 87% (Fig 4.71), higher than 82% (accuracy based on cross validation).

Figure. 4 71 Random Forest Algorithm and Accuracy of the model (Use all six parameters to predict thermal comfort)-
Participant 2
For the second way to test accuracy of the model of using skin temperature, indoor temperature,
EDA, and stress level to predict thermal comfort, we chose first day’s data to train a model,
and chose the last day’s data to test the model. The accuracy is 89% (Fig 4.72), higher than 82%
(accuracy based on cross validation).

Figure. 4 72 Random Forest Algorithm for using first day to train a model and last day data to test accuracy of the model
(Use skin temperature, indoor temperature, EDA, and stress level to predict thermal comfort) - Participant 2

70

In the same way, the accuracy of the random forest model of using all six parameters to predict
thermal comfort of participant 2 is around 89.3% (Fig 4.73), higher than 87% (accuracy based
on cross validation), and higher than 89.7% (use more relevant parameters to predict thermal
comfort-accuracy based on first day and last day data).

Figure. 4 73 Random Forest Algorithm for using first day to train a model and last day data to test accuracy of the model
(Use all six parameters to predict thermal comfort)-Participant 2
Random forest was applied to investigate the importance of all six parameters affecting thermal
comfort, as Fig 4.74 illustrated. Among six parameters, indoor temperature, indoor relative
humidity, EDA, and skin temperature are more important parameters to thermal comfort.

Figure. 4 74 Feature importance (six parameters) graph from Random Forest - Participant 2
The accuracy of the model of using indoor temperature, indoor relative humidity, EDA, and
skin temperature to predict thermal comfort is around 93% based on cross validation (Fig 4.75),
higher than 82% (accuracy based on stepwise linear regression analysis result – cross
validation), and higher than 87% (accuracy of the model of using all six parameters to predict
thermal comfort– cross validation).

71

Figure. 4 75 Random Forest Algorithm and Accuracy of the model (Use indoor temperature, indoor relative humidity, EDA,
and skin temperature to predict thermal comfort, use cross validation to test accuracy)-Participant 2
As for using the second way to test accuracy of the model of using indoor temperature, indoor
relative humidity, EDA, and skin temperature to predict thermal comfort (chose first day’s data
to train a model and chose the last day’s data to test the model). The accuracy is around 92%
(Fig 4.76), lower than 93% (accuracy based on cross validation), and higher than 90%
(accuracy of the model of using all parameters to predict thermal comfort).

Figure. 4 76 Random Forest Algorithm for using first day to train a model and last day data to test accuracy of the model
(Use indoor temperature, indoor relative humidity, EDA, and skin temperature to predict thermal comfort)-Participant 2
Table 4.17 shows the summary of the more important parameters based on stepwise linear
regression analysis result and random forest result.
Table 4. 17 Summary of relevant parameters based on two different methods - Participant 2
(S – based on stepwise linear regression result, R – based on random forest result)
Method More relevant parameters
Stepwise linear regression Skin temperature, indoor temperature, EDA and stress level
Random forest Indoor temperature, indoor relative humidity, EDA, and skin temperature
Table 4.18 shows the summary of accuracy of different thermal comfort prediction model, in
the table, S means the accuracy based on the stepwise linear regression result, R means the
accuracy based on the random forest result.

72

Table 4.18 Summary of accuracy of different thermal comfort prediction model - Participant 2
(S – based on stepwise linear regression result, R – based on random forest result)
Model details and way of testing accuracy Accuracy of the model
S - Use relevant parameters to predict (cross validation) 82%
S - Use relevant parameters to predict (first day and last day) 89%
R - Use relevant parameters to predict (cross validation) 93%
R - Use relevant parameters to predict (first day and last day) 92%
Use all parameters to predict (cross validation) 87%
Use all parameters to predict (first day and last day) 90%
The analysis process of the other ten participants is the same as that of the first two participants.
Please refer to the appendix for detailed analysis process.

73

Chapter 5: RESULTS
5.1 Physiological signal most relevant to IAQ parameters
According to the Pearson Correlation analysis from chapter four, different participants have
different correlations between four bio-signals and IAQ parameters (PM2.5 and CO2). We
concluded some results based on the Pearson Correlation analysis and stepwise linear
regression analysis.
In all the tables in 5.1, numbers 1~12 represent each participant, and #1~4 represent the
sequence numbers from relatively strong to the relatively weak correlation between four bio-
signals and IAQ parameters (for Pearson correlation analysis results and stepwise linear
regression analysis results), # represent most significant, and #4 represent least significant. For
example, in Tale 5.1, the bio-signals in line #1 are the most relevant bio-signal to indoor CO2
concentration for each participant. Furthermore, some Pearson correlation coefficients between
bio-signals and indoor CO2 concentration are the same. For bio-signals with the same
coefficients (for the individual subject), they were marked in red in the table. The two Bio-
signals next to each other are red, indicating that they have the same correlation coefficient
with indoor CO2 concentration. Besides, in all tables, HR represents heart rate, stress represents
stress level, and S_T represents skin temperature.
5.1.1Results based on Pearson Correlation analysis
5.1.1.1 The most correlated bio-signal to indoor CO2 concentration
The rank of correlation between four bio-signals of each participant and indoor CO2
concentration is shown in Table 5.1.
Table 5. 1Rank of correlation between bio-signals and CO2 for each participant
1 2 3 4 5 6 7 8 9 10 11 12
#1 stress EDA S_T HR S_T S_T S_T S_T HR HR HR HR
#2 HR HR HR stress stress EDA HR HR S_T stress EDA stress
#3 EDA S_T EDA S_T EDA HR stress stress EDA S_T S_T S_T
#4 S_T stress stress EDA HR stress EDA EDA stress EDA stress EDA
Table 5.2 shows the number of total times each bio-signal appears in the first rank place.
According to the results, both HR and skin temperature appeared five times in first rank place,
which was the most among the four Bio-signals, proving that heart rate and skin temperature
were the two factors more correlated to indoor CO2 concentration than stress level and EDA.

74

Table 5. 2 Number of total times each bio-signal appears in the first rank (CO2)
Bio-signals Times of rank first
HR 5
stress 3
EDA 1
S_T 5
To further explore which bio-signal is most important for indoor CO2 concentration. We
assumed that each bio-signal placing first rank gets four points, each bio-signal placing second
rank gets three points, the third rank bio-signal gets two points, and bio-Signal placing last rank
gets one point; also, bio-signals, tied for the ranking, received the same score. The score
distribution for each bio-signal is shown in Table 5.3.
Table 5. 3 Score distribution for each bio-signal
1 2 3 4 5 6 7 8 9 10 11 12
#1 4 4 4 4 4 4 4 4 4 4 4 4
#2 3 3 3 3 3 3 3 3 3 4 3 4
#3 3 2 2 2 3 2 2 3 3 2 2 2
#4 1 1 1 2 1 1 1 1 1 1 1 1
The calculated total score for kind of the four bio-signals is shown in Fig 5.1 below, the total
score stress level gets is 28, heart rate gets 38 scores, EDA gets 27 scores, and skin temperature
gets 34 scores. Because heart rate gets the highest score, it is more relevant to indoor CO2
concentration than stress level, EDA, and skin temperature.

Figure 5.1.1 Total scores for each bio-signal – CO2
The experimenters participating in our experiment were evenly distributed between males and
females (6 males and six females), so we divided them into two groups based on gender and
explored the bio-signal most relevant to indoor CO2 concentration. After analysis and
calculation, the total number of times that each type of bio-signal appeared in the first place in
28
38
27
34
0
10
20
30
40
stress HR EDA S_T
Total scores for each bio-signal (CO
2
)

75

the male group is shown in Table 5. 4. The score of each type of bio-signal is shown in Fig 5.2.
Table 5. 4 Number of total times each bio-signal appears in the first rank in male group
Bio-signals Times of rank first
HR 4
stress 2
EDA 0
S_T 2
According to Table 5. 4, heart rate ranks first for the most time, which is 4, and according to
the total score, each kind of bio-signals get shown in Fig. 5.1.2, heart rate also got the highest
score is 22. Therefore, heart rate is also the more relevant bio-signal to indoor CO2
concentration in the male group. Also, EDA is relatively irrelevant.

Figure 5.1.2 Total scores for each bio-signal in male group – CO2
Table 5.5 shows the total number of times that each type of bio-signal appears in the first rank
place in the female group, and the scores of each type of bio-signal are shown in the Fig 5.5.
Table 5. 5 Number of total times each bio-signal appears in the first rank in male group
Bio-signals Times of rank first
HR 1
stress 1
EDA 1
S_T 3
According to Table 5. 5, skin temperature ranks first for most time in the female group, which
is 4, and based on the total score each kind of bio-signals get showing in Fig. 5.1.3, both EDA
and skin temperature get the highest score, which is 18, and stress level gets the lowest score.
Therefore, except for skin temperature, EDA is also the more relevant bio-signal to indoor CO2
concentration than stress level and heart rate in the female group. The results are different from
the female group’s results.
17
22
9
16
0
10
20
30
stress HR EDA S_T
Total scores for each bio-signal in male
group (CO
2
)

76

Figure 5.1.3 Total scores for each bio-signal in female group – CO2
Fig.5.1.4, showing the comparison chart of total scores for each bio-signals in male and female
groups, illustrates that the scores of stress level and heart rate of males are higher than that of
females. The scores of skin temperature and EDA of females are higher than those of males,
testifying that indoor CO2 concentration has a more significant impact on the stress level and
heart rate of males than on that two bio-signals of females and has a more significant impact
on EDA and skin temperature of females than on those two bio-signals of males.

Figure 5.1.4 Comparison chart of total scores for each bio-signals in male and female groups – CO2
5.1.1.2The most correlated bio-signal to indoor PM2.5 concentration
Table 5.6 shows the rank of correlation between four bio-signals of each participant and indoor
PM2.5 concentration.
Table 5. 6 Rank of correlation between bio-signals and PM2.5 for each participant
1 2 3 4 5 6 7 8 9 10 11 12
#1 HR EDA S_T HR S_T S_T EDA HR S_T S_T S_T HR
#2 stress S_T HR EDA stress HR stress stress stress stress HR S_T
#3 EDA HR EDA stress EDA stress HR EDA HR EDA stress stress
#4 S_T stress stress S_T HR EDA S_T S_T EDA HR EDA EDA
11
16
18 18
0
10
20
stress HR EDA S_T
Total scores for each bio-signal in female
group (CO
2
)
0
10
20
stress HR EDA S_T
Comparison chart of total scores for each bio-
signal in male and female groups (CO
2
)
male female

77

Table 5.7 shows the number of total times each bio-signal appears in the first rank. According
to the results, both HR and skin temperature appeared most times in first rank place, indicating
that heart rate and skin temperature were the two factors more correlated to indoor PM2.5
concentration compared with stress level and EDA.
Table 5. 7 Number of total times each bio-signal appears in the first rank (PM2.5)
Bio-signals Times of rank first
HR 6
stress 1
EDA 3
S_T 6
Same as the process of exploring which bio-signal is most related to indoor CO2 concentration,
we also adopted fractional calculation method to further explore the most relevant bio-signal
to indoor PM2.5 concentration. The score distribution for each bio-signal is shown in Table
5.8.
Table 5. 8 Score distribution for each bio-signal
1 2 3 4 5 6 7 8 9 10 11 12
#1 4 4 4 4 4 4 4 4 4 4 4 4
#2 3 3 3 4 3 4 3 3 3 4 4 3
#3 3 2 2 2 2 2 2 2 2 3 2 3
#4 1 1 1 1 1 1 1 2 1 2 1 1
The calculated total score for kind of the four bio-signals is shown in the Fig 5.1.5. The total
score stress level gets is 30, heart rate gets 36, EDA gets 28, and skin temperature gets 35.
Because heart rate gets the highest score, it is more relevant to indoor CO2 concentration
compared with stress level, EDA, and skin temperature.

Figure 5.1.5 Total score for each bio-signal – PM2.5
We divided all the participants based on gender and explored the bio-signal most relevant to
30
36
28
35
0
10
20
30
40
stress HR EDA S_T
Total scores for each bio-signal (PM2.5)

78

indoor PM2.5 concentration in two groups. After analysis and calculation, the total number of
times that each type of bio-signal appeared in the first place in the male group is shown in
Table 5. 9.
Table 5. 9 Number of total times each bio-signal appears in the first rank in male group (PM2.5)
Bio-signals Times of rank first
HR 4
stress 1
EDA 2
S_T 2
As Table 5.1.9 illustrates, heart rate rank first for most time, which is 4, and according to the
total score each kind of bio-signals get showing in Fig. 5.1.6, heart rate also got the highest
score, 19. Therefore, heart rate is also the more relevant bio-signal to indoor PM2.5
concentration in male group.

Figure 5.1.6 Total scores for each bio-signal in male group – PM2.5
Table 5.10 shows the total number of times that each type of bio-signal appears in the first rank
place in the female group, and the scores of each type of bio-signal are shown in the Fig 5.6.
Table 5. 10 Number of total times each bio-signal appears in the first rank in female group
Bio-signals Times of rank first
HR 2
stress 0
EDA 1
S_T 4
As Table 5.10 illustrates, skin temperature ranks first for most time in the female group, which
is 4, and based on the total score each kind of bio-signals gets in Fig. 5.1.7, skin temperature
gets the highest score is 20. Therefore, skin temperature is the more relevant bio-signal to
indoor PM2.5 concentration than heart rate, stress level, and EDA in the female group.
However, the results are different from the male group’s results.
17
19
14
15
0
10
20
stress HR EDA S_T
Total scores for each bio-signal in male group
(PM2.5 )

79

Figure 5.1.7 Total scores for each bio-signal in female group – PM2.5
Fig 5.1.8, showing the comparison chart of total scores for each bio-signals in male and female
groups, illustrates that the scores of stress level and heart rate of males are higher than that of
females. The scores of skin temperature of females are higher than that of males, indicating
that indoor PM2.5 concentration has a more significant impact on the stress level and heart rate
of males than on that two bio-signals of females and has a greater impact on the skin
temperature of females than on the skin temperature of males.

Figure 5.1.8 Comparison chart of total scores for each bio-signals in male and female groups – PM2.5
5.1.2 Results based on Stepwise Linear Regression
5.1.2.1 The most significant bio-signal to indoor CO2 concentration
In stepwise linear regression results, it only showed the significant relevant bio-signals, and
the non-significant bio-signals to IAQ parameters did not show in the results sheet of all
participants. Therefore, some grids are blank for tables about the rank of correlation and the
13
16
14
20
0
10
20
stress HR EDA S_T
Total scores for each bio-signal in female
group (PM2.5 )
0
10
20
stress HR EDA S_T
Comparison chart of total scores for each bio-signal in
two groups (PM 2.5)
male female

80

scores.
The rank of the significance of each four bio-signals of each participant to indoor CO2
concentration is shown in Table 5.11.
Table 5. 11 Rank of significance of four bio-signals to indoor CO2 concentration
1 2 3 4 5 6 7 8 9 10 11 12
#1 stress stress S_T HR S_T S_T S_T S_T HR HR HR HR
#2 EDA HR HR EDA HR HR EDA S_T EDA S_T
#3 EDA EDA HR stress stress stress stress
#4 S_T
Table 5.12 shows the number of total times each bio-signal appears in the first rank place. Both
HR and skin temperature appeared five times in first rank place, which was largest among the
four Bio-signals, indicating that heart rate and skin temperature were the two factors more
significantly related to indoor CO2 concentration compared with stress level and EDA; the
results are the same as the results from Pearson correlation coefficient analysis.
Table 5. 12 Number of total times each bio-signal appears in the first rank (CO2)
Bio-signals Times of rank first
HR 5
stress 2
EDA 0
S_T 5
Same as the process in 5.1.1, we also adopted the fractional calculation method in 5.1.2 to
explore the most relevant bio-signals to IAQ parameters (each bio-signal placing first rank gets
four points, each bio-signal placing the second rank gets three points, the third rank bio-signal
gets two points, and bio-signal placing last rank gets one point). The score distribution for each
bio-signal is shown in Table 5.13.
Table 5. 13 Score distribution for each bio-signal (CO2)
1 2 3 4 5 6 7 8 9 10 11 12
#1 4 4 4 4 4 4 4 4 4 4 4 4
#2 3 3 3 3 3 3 3 3 3 3
#3 2 2 2 2 2 2 2
#4 1
The calculated total score for four bio-signals is shown in the Fig 5.1.9 below. The total score
stress level gets is 16, heart rate gets 34, EDA gets 16, and skin temperature gets 27. Obviously,
heart rate gets the highest score, it is more relevant to indoor CO2 concentration compared with
stress level, EDA, and skin temperature.

81

Figure 5.1.9 Total scores for each bio-signal – CO2
After analysis and calculation, the total number of times that each type of bio-signal ranks first
in the male group is shown in Table 5.14.
Table 5.14 Number of total times each bio-signal appears in the first rank in male group (CO2)
Bio-signals Times of rank first
HR 4
stress 0
EDA 0
S_T 2
According to Table 5.14, heart rate rank first for most time, and based on the scores showing
on Fig. 5.1.10, heart rate also got 19, which is the highest score. Therefore, heart rate is also
the more relevant bio-signal to indoor CO2 concentration in male group.

Figure 5.1.10 Total scores for each bio-signal in male group – CO2
Table 5.15 shows the total number of times that each type of bio-signal appears in the first rank
place in the female group.

16
34
16
27
0
10
20
30
40
stress HR EDA S_T
Total scores for each bio-signal (CO
2
)
4
19
3
14
0
10
20
stress HR EDA S_T
Total scores for each bio-signal in male group
(CO
2
)

82

Table 5. 15 Number of total times each bio-signal appears in the first rank in female group (CO2)
Bio-signals Times of rank first
HR 1
stress 2
EDA 0
S_T 3
According to Table 5.1.15, skin temperature ranks first for most time in female group and based
on the total score each kind of bio-signals get showing in Fig. 5.9; even skin temperature ranks
first most time, the score heart rate gets is largest. Therefore, heart rate is also the more relevant
bio-signal to indoor CO2 concentration than stress level, EDA, and skin temperature in the
female group. Besides, this result of the female group is different from the results based on
Pearson correlation analysis in 5.1.11.

Figure 5.1.11 Total scores for each bio-signal in female group – CO2
As 5.1.12 illustrates, the scores of heart rate and skin temperature of males are higher than that
of females. Also, the scores of stress level of females are higher than that of males, indicating
that indoor CO2 concentration has a greater impact on the heart rate and skin temperature of
males than on that two bio-signals of females, and has a greater impact on the stress level and
EDA of females than on those bio-signals of males.

Figure 5.1.12 Comparison chart of total scores for each bio-signals in male and female groups – CO2
12
15
13 13
0
10
20
stress HR EDA S_T
Total scores for each bio-signal in female
group (CO
2
)
0
10
20
stress HR EDA S_T
Comparison chart of total scores for each bio-signal in
male and female groups (CO
2
)
male female

83

5.1.2.2 The most significant bio-signal to indoor PM2.5 concentration
The rank of significance between four bio-signals of each participant to indoor PM2.5
concentration is shown in Table 5.16.
Table 5. 16 Rank of significance of four bio-signals to indoor PM2.5 concentration
1 2 3 4 5 6 7 8 9 10 11 12
#1 HR EDA S_T HR S_T HR EDA HR S_T stress S_T HR
#2 EDA HR HR EDA HR S_T S_T stress HR HR S_T
#3 EDA EDA EDA S_T stress EDA
#4 stress stress
Table 5.17 shows the number of total times each bio-signal appears in the first rank place.
According to the results, heart rate appeared most times in first rank place, indicating that heart
rate is more significantly related to indoor PM2.5 concentration compared with stress level,
EDA and skin temperature.
Table 5. 17 Number of total times each bio-signal appears in the first rank (PM2.5)
Bio-signals Times of rank first
HR 5
stress 1
EDA 2
S_T 4
The score distribution for each bio-signal based on stepwise linear regression analysis is shown
in Table 5.18.
Table 5. 18 Score distribution for each bio-signal
1 2 3 4 5 6 7 8 9 10 11 12
#1 4 4 4 4 4 4 4 4 4 4 4 4
#2 3 3 3 3 3 3 3 3 3 3 3
#3 2 2 2 2 2 2
#4 1 1
The calculated total score for kind of the four bio-signals is shown in the Fig 5.1.13, heart rate
gets the highest score, and stress level gets the lowest score. Hence, heart rate is more relevant
to indoor CO2 concentration compared with stress level, EDA, and skin temperature.

84

Figure 5.1.13 Total scores for each bio-signal – PM2.5
After analysis and calculation, the total number of times that each type of bio-signal appeared
in the first place in the male group is shown in Table 5.19
Table 5. 19 Number of total times each bio-signal appears in the first rank in male group (PM2.5)
Bio-signals Times of rank first
HR 3
stress 1
EDA 1
S_T 1
According to the total score each kind of bio-signal gets showing in Fig. 5.1.14, heart rate also
got the highest score. Therefore, heart rate is also the more relevant bio-signal to indoor PM2.5
concentration in male group.

Figure 5.1.14 Total score for each bio-signal in male group – PM2.5
Table 5.20 shows the total number of times that each type of bio-signal appears in the first rank
11
35
22
27
0
10
20
30
40
stress HR EDA S_T
Total scores for each bio-signal (PM2.5)
7
18
9
12
0
10
20
stress HR EDA S_T
Total scores for each bio-signal in male group (PM2.5 )

85

place in the female group.
Table 5. 20 Number of total times each bio-signal appears in the first rank in female group (PM2.5)
Bio-signals Times of rank first
HR 2
stress 0
EDA 1
S_T 3
According to Fig 5.1.15, heart rate gets the highest score, which is 17, and stress level gets the
lowest score. Therefore, heart rate is more relevant to indoor PM2.5 concentration compared
with stress level, EDA, and skin temperature in female group. Besides, this result of female
group is different from the results based on Pearson correlation analysis in 5.1.

Figure 5.1.15 Total scores for each bio-signal in female group – PM2.5
As Fig 5.1.16 illustrates, the scores of stress level and heart rate of males are higher than that
of females. The scores of skin temperature and EDA of females are higher than that of males,
indicating that indoor PM2.5 concentration has a greater impact on the stress level and heart
rate of males than on that two bio-signals of females, and has a greater impact on the skin
temperature and EDA of females than on those two bio signals of males. In addition, heart rate
scored the highest of the four bio signals for both male and female groups. Besides, except
EDA, the results concluded from stepwise linear regression are the same as the results
concluded from Pearson correlation coefficients.
4
17
13
15
0
10
20
stress HR EDA S_T
Total scores for each bio-signal in female group
(PM2.5 )

86

Figure 5.1.16 Comparison chart of total scores for each bio-signals in male and female groups – PM2.5
5.2 More relevant physiological signals to IAQ Evaluation
5.2.1 Results based on Stepwise Linear Regression
According to the stepwise analysis results about exploring relevant physiological signals to
IAQ Evaluation in chapter four, the different participant has different results. In this part, we
summarized all participants’ bio-signals that significantly related to IAQ Evaluation and sorted
them. In addition, we summarized the two kinds of accuracy, one accuracy is based on cross
validation (use 80% data to train the model, used 20% data to test the model); the other accuracy
is based on using first day data to train a model and used the last day data to test the model)
In all the tables shown in 5.2 and 5.3, numbers 1~12 represent each participant, and #1~4
represent the sequence numbers from most significant to least significant. For example, in Tale
5.2.1, the bio-signals located in line #1 are most significantly related to IEQ Evaluation for
each participant, and the bio-signals placed in row #3 are least significantly related to IEQ
Evaluation. Also, in all tables, HR represents heart rate, stress represents stress level, and S_T
represents skin temperature.
Tale 5.2.1 shows the rank of significance of important bio-signals of each participant to IEQ
Evaluation. In this table, A1 represents the accuracy based on cross validation. A2 represents
the accuracy based on using first day data to train the model and last day data to test the model,
A1 and A2 are for the model accuracy of using relevant bio-signals to predict IAQ Evaluation.
A1-a and A2-a are representing the model accuracy of using all bio-signals to predict IAQ
Evaluation.

0
10
20
stress HR EDA S_T
Comparison chart of total scores for each bio-signal in
male and female groups (PM2.5)
male female

87

Table 5. 21 Rank of significance of important bio-signals of each participant to IEQ Evaluation - Stepwise
1 2 3 4 5 6 7 8 9 10 11 12
#1 stress HR stress EDA stress stress EDA S_T HR HR stress HR
#2 S_T HR HR EDA HR S_T stress S_T
#3 S_T stress
#4 EDA
A1 90% 96.6% 84% 96% 98% 87% 93% 56% 79% 63% 72% 61%
A2 93% 96.6% 91% 93% 97% 90% 94% 34% 82% 62% 86% 77%
A1-a 89% 96.4% 84% 95% 97% 94% 90% 52% 71% 69% 79% 65%
A2-a 97% 96.7% 91% 98% 98% 97% 90% 80% 88% 89% 87% 81%
According to the accuracy showing in table 5.2.1, in most cases, the accuracy of using first data
and last data is higher than the accuracy based on cross validation. Only four situations (red
font) showed the opposite condition.
Table 5.22 shows the number of total times each bio-signal appears in the first rank place.
Stress level appears most times in first rank place, indicating that stress level was more
significantly related to IAQ Evaluation compared with heart rate, skin temperature and EDA.
Table 5. 22 Number of total times each bio-signal appears in the first rank (IAQ Evaluation)
Bio-signals Times of rank first
HR 4
stress 5
EDA 2
S_T 1
Same as the process in 5.1, we adopted fractional calculation method in 5.2 to explore the most
relevant bio-signals to IEQ Evaluation (each bio-signal placing first rank gets four points, each
bio-signal placing second rank gets three points, the third rank bio-signal get two points, and
the last rank bio-signal get 1 point). The score distribution for each bio-signal is shown in Table
5.23.
Table 5. 23 Score distribution for each bio-signal (IEQ Evaluation)
1 2 3 4 5 6 7 8 9 10 11 12
#1 4 4 4 4 4 4 4 4 4 4 4 4
#2 3 3 3 3 3 3 3 3
#3 2 2
#4 1
The calculated total score for four kind of bio-signals is shown in the Fig 5.2.1 below, stress
level and heart rate get the highest score, indicating that heart rate and stress level are more
relevant to IEQ Evaluation compared with other two bio-signals.

88

Figure 5.2.1 Total scores for each bio-signal – IAQ Evaluation
We divided 12 participants into two groups based on gender and explored the bio-signal most
relevant to IEQ Evaluation. The total number of times that each type of bio-signal ranks first
in the male group is shown in Table 5.24.
Table 5. 24 Number of total times each bio-signal appears in the first rank in male group (IAQ Evaluation)
Bio-signals Times of rank first
HR 2
stress 1
EDA 2
S_T 1
Based on the scores showing on Fig. 5.2.2, heart rate also gets 11, which is the highest score,
illustrating that heart rate is more significant bio-signal to IAQ Evaluation in male group.

Figure 5.2.2 Total scores for each bio-signal in male group –IAQ Evaluation
Table 5.25 shows the total number of times that each type of bio-signal appears in the first rank
place in the female group.
6
11
8
10
0
4
8
12
stress HR EDA S_T
Total scores for each bio-signal in male
group (IAQ Evaluation)

89

Table 5. 25 Number of total times each bio-signal ranks first in female group (IAQ Evaluation)
Bio-signals Times of rank first
HR 2
stress 4
EDA 0
S_T 0
According to Table 5.25, stress level ranks first for most time in female group, which is
different from males’ group. Also, Fig 5.2.3 shows total score each kind of bio-signals get, and
stress level get the highest score, which is also different from the results showing in males’
group. Therefore, stress level is more significant bio-signal to IAQ Evaluation in female group.

Figure 5.2.3 Total scores for each bio-signal in female group – IAQ Evaluation
As Fig 5.2.4 illustrates, the total scores of stress level and heart rate in female group are higher
than them in male group. Also, the scores of EDA and skin temperature of females are higher
than those two bio-signals of males.

Figure 5.2.4 Comparison chart of total scores for each bio-signals in male and female groups – IEQ Evaluation

90

5.2.2 Results based on Random Forest result
All participants’ bio-signals that significantly related to IAQ Evaluation were also summarized
and sorted based on random forest results showing in chapter 4. In the same way, we
summarized the two kinds of accuracy, one accuracy is based on cross validation (use 80%
data to train the model, used 20% data to test the model); the other accuracy is based on using
first day data to train a model and used the last day data to test the model).
Tale 5.2.1 shows the rank of significance of important bio-signals of each participant to IEQ
Evaluation based on random forest result. In this table, A1 represents the accuracy based on
cross validation. A2 represents the accuracy based on using first day data to train the model
and last day data to test the model, A1 and A2 are for the model accuracy of using relevant bio-
signals to predict IAQ Evaluation. A1-a and A2-a are representing the model accuracy of using
all bio-signals to predict IAQ Evaluation.
Table 5. 26 Rank of significance of important bio-signals of each participant to IEQ Evaluation – Random Forest
1 2 3 4 5 6 7 8 9 10 11 12
#1 HR EDA stress EDA EDA stress EDA EDA HR EDA S_T EDA
#2 stress S_T S_T HR S_T S_T stress EDA stress EDA S_T
A1 83% 96.7% 81% 96% 97% 94% 93% 61% 77% 74% 61% 61%
A2 92% 96.2% 86% 93% 97% 93% 94% 57% 86% 76% 77% 72%
A1-a 89% 96.4% 84% 95% 97% 94% 90% 52% 71% 69% 79% 65%
A2-a 97% 96.7% 91% 98% 98% 97% 90% 80% 88% 89% 87% 81%
As the accuracy showing in table 5.26, in most cases, the accuracy of using first data and last
data is higher than the accuracy based on cross validation. Only four situations (red font)
showed the opposite condition.
Table 5.27 shows the number of total times each bio-signal appears in the first rank place. EDA
appears most times in first rank place, which is different from the results based on stepwise
linear regression.
Table 5. 27 Number of total times each bio-signal appears in the first rank (IAQ Evaluation)
Bio-signals Times of rank first
HR 2
stress 1
EDA 8
S_T 1
Same as the process in 5.1, we adopted fractional calculation method in 5.2 to explore the most
relevant bio-signals to IEQ Evaluation (each bio-signal placing first rank gets two points, each
bio-signal placing second rank gets one points). The score distribution for each bio-signal is

91

shown in Table 5.28.
Table 5. 28 Score distribution for each bio-signal (IEQ Evaluation)
1 2 3 4 5 6 7 8 9 10 11 12
#1 2 2 2 2 2 2 2 2 2 2 2 2
#2 1 1 1 1 1 1 1 1 1 1 1
The calculated total score for four kind of bio-signals is shown in the Fig 5.2.5 below, EDA
get the highest score.

Figure 5.2.5 Total scores for each bio-signal – IEQ Evaluation
5.3 More relevant factors to IAQ Evaluation
5.3.1 Results based on Stepwise linear regression
According to the analysis results about exploring relevant parameters (between four bio-signals,
indoor temperature, indoor relative humidity and two IAQ parameters) to IAQ Evaluation in
4.4, we summarized first four parameters that significantly related to IAQ Evaluation and
sorted them. Moreover, we also summarized the two kinds of accuracy.
Tale 5.29 shows the rank of significance of important parameters to IEQ Evaluation and the
accuracy of the model of predicting IAQ Evaluation of each participant based on random forest
result. In this table, A1 represents the accuracy based on cross validation. A2 represents the
accuracy based on using first day data to train the model and last day data to test the model, A1
and A2 are for the model accuracy of using relevant bio-signals to predict IAQ Evaluation. A1-
a and A2-a are representing the model accuracy of using all bio-signals to predict IAQ
Evaluation.
In all tables, HR represents heart rate; stress represents stress level, S_T represents skin
temperature, RH% represent indoor relative humidity, and Temp represents indoor temperature.

92

Tale 5.29 shows the rank of significance of first significant parameters of each subject to IEQ
Evaluation.
Table 5. 29 Rank of significance of significant parameters to IEQ Evaluation and the prediction model accuracy of each
subject – Stepwise linear regression
1 2 3 4 5 6 7 8 9 10 11 12
#1 PM2.5 Temp RH% EDA stress stress RH% CO2 CO2 CO2 RH% PM2.5
#2 stress PM2.5 stress HR RH% Temp S_T RH% HR RH% stress HR
#3 CO2 RH% S_T Temp RH% stress Temp stress PM2.5 PM2.5 Temp
#4 Temp EDA Temp RH% PM2.5 PM2.5 PM2.5 PM2.5 Temp EDA
A1 91% 98% 90% 98% 97% 98% 90% 93% 86% 91% 78% 77%
A2 96% 99% 95% 100% 94% 98% 98% 93% 91% 95% 82% 87%
A1-a 89% 99% 94% 95% 98% 94% 93% 88% 87% 89% 95% 72%
A2-a 99% 98% 99% 98% 98% 99% 99% 95% 96% 98% 93% 93%
Table 5.30 shows the number of total times eight parameters appear in the first rank. Indoor
relative humidity and indoor CO2 concentration appear most times in first rank place.
Table 5.30 Number of total times each eight parameters appears in the first rank (IAQ Evaluation)
Bio-signals Times of rank first
Temp 1
RH% 3
PM2.5 2
CO2 3
HR 0
stress 2
EDA 1
S_T 0
Same as the process in 5.1 and 5.2, we adopted fractional calculation method in 5.3 to explore
the most relevant parameters to IEQ Evaluation (each parameters placing first rank get four
points, the second rank parameters get three points, the third rank parameters get two points,
and the last rank parameters get 1 point). The score distribution for each parameter is shown in
Table 5.31.
Table 5. 31 Score distribution for each eight parameters (IEQ Evaluation)
1 2 3 4 5 6 7 8 9 10 11 12
#1 4 4 4 4 4 4 4 4 4 4 4 4
#2 3 3 3 3 3 3 3 3 3 3 3 3
#3 2 2 2 2 2 2 2 2 2 2 2
#4 1 1 1 1 1 1 1 1 1 1
The calculated total score for each eight parameters is shown in the Fig 5.19 below, indoor

93

relative humidity gets the highest score, followed by stress level, PM2.5, indoor temperature,
indoor CO2 concentration, heart rate, EDA, and skin temperature. Therefore, indoor relative
humidity is more relevant to IAQ Evaluation compared with other seven parameters, and stress
level is the most relevant bio signal to IAQ Evaluation compared with other three bio-signals.

Figure 5.3.1 Total scores for each eight parameters – IAQ Evaluation
5.3.2 Results based on Random Forest result
We also explore the relevant parameters (between four bio-signals, indoor temperature, indoor
relative humidity and two IAQ parameters) to IAQ Evaluation in 4.4 based on random forest
result, we summarized first four parameters that significantly related to IAQ Evaluation and
sorted them. Same as 5.3.4, we also summarized the two kinds of accuracy.
Tale 5.3.4 shows the rank of significance of important parameters to IEQ Evaluation and the
accuracy of the model of predicting IAQ Evaluation of each participant based on random forest
result. In this table, A1 represents the accuracy based on cross validation. A2 represents the
accuracy based on using first day data to train the model and last day data to test the model, A1
and A2 are for the model accuracy of using relevant bio-signals to predict IAQ Evaluation. A1-
a and A2-a are representing the model accuracy of using all bio-signals to predict IAQ
Evaluation.
In all tables, HR represents heart rate; stress represents stress level, S_T represents skin
temperature, RH% represent indoor relative humidity, and Temp represents indoor temperature.
Tale 5.32 shows the rank of significance of first significant parameters of each subject to IEQ
Evaluation.

94

Table 5. 32 Rank of significance of significant parameters to IEQ Evaluation and the model accuracy of each subject –
Random Forest
1 2 3 4 5 6 7 8 9 10 11 12
#1 PM2.5 EDA RH% RH% CO2 Temp S_T Temp CO2 Temp CO2 RH%
#2 stress CO2 PM2.5 EDA HR PM2.5 RH% CO2 EDA CO2 RH% EDA
#3 CO2 Temp CO2 CO2 PM2.5 HR EDA PM2.5 RH% PM2.5 Temp Temp
#4 HR PM2.5 S_T Temp Temp stress CO2 RH% S_T RH% PM2.5 S_T
A1 86% 99% 98% 100% 98% 98% 90% 91% 82% 91% 83% 75%
A2 98% 98% 99% 100% 97% 98% 98% 93% 96% 92% 94% 89%
A1-a 89% 99% 94% 95% 98% 94% 93% 88% 87% 89% 95% 72%
A2-a 99% 98% 99% 98% 98% 99% 99% 95% 96% 98% 93% 93%
Table 5.33 shows the number of total times eight parameters appear in the first rank. Indoor
temperature, indoor relative humidity and indoor CO2 concentration appear most times in first
rank place.
Table 5.33 Number of total times each eight parameters appears in the first rank (IAQ Evaluation)
Bio-signals Times of rank first
Temp 3
RH% 3
PM2.5 1
CO2 3
HR 0
stress 0
EDA 1
S_T 1
We adopted fractional calculation method in 5.3 to explore the most relevant parameters to
IEQ Evaluation (each parameters placing first rank get four points, the second rank parameters
get three points, the third rank parameters get two points, and the last rank parameters get 1
point). The score distribution for each parameter based on random forest is shown in Table
5.34.
Table 5.34 Score distribution for each eight parameters (IEQ Evaluation)
1 2 3 4 5 6 7 8 9 10 11 12
#1 4 4 4 4 4 4 4 4 4 4 4 4
#2 3 3 3 3 3 3 3 3 3 3 3 3
#3 2 2 2 2 2 2 2 2 2 2 2
#4 1 1 1 1 1 1 1 1 1 1
The calculated total score for each eight parameters is shown in the Fig 5.3.2 below, indoor
CO2 gets the highest score, followed by indoor relative humidity, indoor temperature, indoor
PM2.5, EDA, skin temperature, heart rate and stress level. Therefore, indoor CO2 is more

95

relevant to IAQ Evaluation compared with other seven parameters based on random forest
analysis, and EDA is the most relevant bio signal to IAQ Evaluation compared with other three
bio-signals based on random forest analysis.

Figure 5.3.2 Total scores for each eight parameters – IAQ Evaluation
5.4 Some specific relationship about two bio-signals (HR and stress level) and IAQ
As 5.1concludes, except for the results from the female group based on Pearson's analysis,
showing indoor PM2.5 concentration had a greater impact on skin temperature, and indoor
CO2 concentration had a greater impact on EDA and skin temperature, all the other results
showed that indoor PM2.5 concentration and indoor CO2 concentration had the greatest impact
on heart rate. Therefore, we concluded that heart rate is the most relevant bio-signal to IAQ
parameters compared with stress level, EDA, and skin temperature.
As 5.2 concludes, except female groups show stress level is more relevant to IAQ Evaluation
among all four bio-signals, the results from the whole group and female group illustrate that
heart rate is more relevant to IAQ Evaluation than the other three bio signals. Besides, as 5.3
concludes, stress level is more significantly related to IAQ Evaluation than the other seven
parameters.
In conclusion, heart rate and stress level are more relevant to IAQ Evaluation based on stepwise
linear regression.
5.4.1 The relationship between two bio-signals (HR and stress level) and IAQ
parameters
5.4.1.1 The relationship between heart rate and IAQ parameters
Table 5.34 shows the Pearson correlation coefficient between heart rate and IAQ parameters

96

for each participant.
Table 5. 34 Pearson correlation coefficient between HR and IAQ parameters for each participant
1 2 3 4 5 6 7 8 9 10 11 12
CO2 0.1 -0.3 0.4 0.4 -0.03 -0.06 -0.06 0.2 -0.3 -0.2 0.2 -0.3
PM2.5 -0.4 -0.2 0.5 0.1 -0.05 -0.1 0.005 0.3 0.3 -0.04 0.3 0.3
Based on Table 5.34, Table 5.35 summarized the total number of positive and negative
correlations between heart rate and IAQ parameters, five subjects’ heart rate are positively
related with indoor CO2 concentration, and seven subjects’ heart rate are negatively correlated
with indoor CO2 concentration. Furthermore, seven participants’ heart rate are positively
correlated with indoor PM2.5 concentration; and five participants’ heart rate is negatively
correlated with indoor PM2.5 concentration.
Table 5. 35 Number of positive and negative correlations between HR and IAQ parameters

CO2 PM2.5
positive 5 7
negative 7 5
5.4.1.2 The relationship between stress level and IAQ parameters
Table 5.36 shows the Pearson correlation coefficient between stress level and IAQ parameters
for each participant.
Table 5.36 Pearson correlation coefficient between stress level and IAQ parameters for each participant
1 2 3 4 5 6 7 8 9 10 11 12
CO2 0.2 -0.01 0.1 0.3 0.2 -0.01 0.03 0.2 -0.08 -0.2 -0.04 -0.3
PM2.5 -0.3 -0.07 0.2 0.09 -0.2 -0.09 -0.03 0.2 0.4 0.1 -0.1 0.2
Based on Table 5.36, Table 5.37 summarized the total number of positive and negative
correlations between heart rate and IAQ parameters, six subjects’ stress levels are positively
correlated with indoor CO2 concentration, and six subjects’ stress levels are negatively
correlated with indoor CO2 concentration. In addition, six participants’ heart rate are positively
correlated with indoor PM2.5 concentration; and six participants’ heart rate is negatively
correlated with indoor PM2.5 concentration.
Table 5. 37 Number of positive and negative correlations between stress level and IAQ parameters

CO2 PM2.5
positive 6 6
negative 6 6

97

5.4.2 The relationship between two bio-signals (HR and stress level) and IAQ
Evaluation
5.4.2.1 The relationship between heart rate and IAQ Evaluation
Table 5.38 shows the Pearson correlation coefficient between heart rate and IAQ Evaluation
for each participant.
Table 5. 38 Pearson correlation coefficient between two bio-signals (HR and stress level) and IEQ Evaluation for each
participant
1 2 3 4 5 6 7 8 9 10 11 12
HR -0.3 0.09 -0.2 0.3 0.05 0.2 0.2 -0.08 0.4 0.2 -0.03 -0.08
stress -0.4 0.06 -0.2 0.1 0.2 0.3 0.3 -0.1 -0.02 0.2 0.2 -0.1
Based on Table 5.39, Table 5.4.6 summarized the total number of positive and negative
correlations between two bio-signals including heart rate and stress level and IAQ Evaluation;
most (seven) subjects’ heart rates are positively correlated with IAQ Evaluation, indicating that
the higher the heart rate, the more satisfied these participants are with IAQ. In addition, most
(seven) participants’ stress levels are positively correlated with IAQ Evaluation, meaning that
the higher the stress level, the more satisfied these participants are with IAQ.
Table 5.39 Number of positive and negative correlations between two bio-signals (HR and stress level) and IEQ Evaluation

HR stress
positive 7 7
negative 5 5
5.5 Most relevant factors to thermal comfort
5.5.1 Results based on Stepwise linear regression
For exploring relevant parameters (between four bio-signals, indoor temperature, and indoor
relative humidity) to thermal comfort and the accuracy of the model of predicting thermal
comfort of each participant based on stepwise linear regression, we summarized the significant
parameters that significantly related to thermal comfort and sorted them based on stepwise
linear regression illustrated in Table 5.5.1. Moreover, we also summarized the two kinds of
accuracy.
In all tables, HR represents heart rate; stress represents stress level, S_T represents skin
temperature, RH% represent indoor relative humidity, and Temp represents indoor temperature.
A1 represents the accuracy based on cross validation. A2 represents the accuracy based on

98

using first day data to train the model and last day data to test the model, A1 and A2 are for the
model accuracy of using relevant parameters to predict thermal comfort. A1-a and A2-a are
representing the model accuracy of using all parameters to predict thermal comfort.
Tale 5.39 shows the rank of significance of significant parameters to thermal comfort and
accuracy of model for predicting thermal comfort of each subject.
Table 5.39 Rank of significance of significant parameters of each subject to thermal comfort - Stepwise
1 2 3 4 5 6 7 8 9 10 11 12
#1 Temp S_T stress Temp HR stress S_T S_T HR Temp RH% S_T
#2 stress Temp S_T HR RH% HR RH% RH% RH% Temp Temp
#3 EDA EDA HR EDA stress Temp S_T EDA
#4 S_T stress Temp RH% HR S_T stress RH%
A1 88% 82% 93% 84% 87% 79% 78% 69% 83% 68% 95% 79%
A2 73% 87% 97% 79% 89% 85% 81% 92% 80% 83% 92% 78%
A1-a 92% 89% 94% 88% 90% 80% 83% 75% 88% 52% 99% 87%
A2-a 95% 90% 96% 90% 90% 91% 87% 91% 87% 94% 98% 85%
Table 5.40 shows the number of total times six parameters appear in the first rank. Based on
the results, skin temperature appears most times in first rank place, indicating that skin
temperature is more significantly related to thermal comfort compared with other five
parameters.
Table 5. 40 Number of total times each eight parameters appears in the first rank (IAQ Evaluation)
Bio-signals Times of rank first
Temp 3
RH% 1
HR 2
stress 2
EDA 0
S_T 4
We adopted the fractional calculation method in 5.41 to explore the most relevant parameters
to thermal comfort (each parameter placing the first rank gets four points, the second rank
parameters get three points, the third rank parameters get two points, and the last rank
parameters get 1 point). The score distribution for each parameter is shown in Table 5.41.
Table 5. 41 Score distribution for each eight parameters (thermal comfort)
1 2 3 4 5 6 7 8 9 10 11 12
#1 4 4 4 4 4 4 4 4 4 4 4 4
#2 3 3 3 3 3 3 3 3 3 3 3
#3 2 2 2 2 2 2 2 2
#4 1 1 1 1 1 1 1 1

99

The calculated total scores for every six parameters are shown in Fig 5.5.1 below; indoor
temperature gets the highest score, followed by skin temperature, indoor relative humidity,
heart rate, stress level, and EDA. Hence, the indoor temperature is more relevant to thermal
comfort than the other five parameters, and skin temperature is the most relevant bio signal to
thermal comfort compared with the other three bio-signals.

Figure 5.5.1 Total scores for each eight parameter – thermal comfort
5.5.2 Results based on random forest
In same way as 5.5.1, we investigate relevant parameters (between four bio-signals, indoor
temperature, and indoor relative humidity) to thermal comfort and the accuracy of the model
of predicting thermal comfort of each participant based on random forest.
Table 5. 42 Rank of significance of significant parameters to thermal comfort and the model accuracy of each subject –
Random Forest
1 2 3 4 5 6 7 8 9 10 11 12
#1 Temp Temp RH% RH% RH% Temp RH% RH% RH% S_T RH% RH%
#2 S_T RH% S_T Temp Temp RH% Temp S_T HR RH% Temp EDA
#3 RH% EDA Temp stress S_T HR EDA Temp Temp stress S_T S_T
#4 EDA S_T EDA S_T stress EDA S_T stress EDA Temp EDA Temp
A1 80% 93% 93% 84% 89% 85% 88% 76% 77% 83% 96% 81%
A2 84% 92% 97% 89% 91% 92% 87% 86% 88% 86% 99% 87%
A1-a 92% 89% 94% 88% 90% 80% 83% 75% 88% 52% 99% 87%
A2-a 95% 90% 96% 90% 90% 91% 87% 91% 87% 94% 98% 85%
The calculated total scores for every six parameters based on random forest are shown in Fig
5.5.2 below; indoor relative humidity gets the highest score, followed by indoor temperature,
skin temperature, EDA, stress level and heart rate. Hence, the indoor relative humidity is more
relevant to thermal comfort than the other five parameters, and skin temperature is the most

100

relevant bio signal to thermal comfort compared with the other three bio-signals.

Figure 5.5.2 Total scores for each six parameter – thermal comfort

32
43
5 6
12
22
0
10
20
30
40
50
Temp RH% HR stress EDA S_T
Total scores for each parameter (thermal comfort) - Random
Forest

101

Chapter 6: DISCUSSION, LIMITATION AND FUTURE WORK
6.1 Conclusion and discussion
In this study, we investigated the correlation between IAQ parameters (PM2.5 and CO2) and
four human bio-signals (heart rate, stress level, EDA, and skin temperature), the association
between four human bio-signals and IAQ Evaluation, and the relationship between four bio-
signals and thermal comfort for twelve participants in their rooms. We concluded that beads
on Pearson correlation analysis, heart rate is the more relevant bio-signal to indoor PM 2.5
concentration and indoor CO2 concentration; the higher the CO2, the lower the heart rate for
the most participants; the higher the indoor PM2.5 concentration, the higher the heart rate.
Based on Pearson correlation and stepwise analysis results shown on chapter 4.1 and 4.2, heart
rate and stress level are two more relevant bio-signals to IAQ Evaluations; the higher the stress
level and heart rate, the more satisfied most participants are with IAQ
As the results of stepwise linear regression and random forest illustrated in chapter 4.3, 4.4. 4.5
and whole chapter 5, the results are different under different analysis method. And different
people have different results, which means we need individually different customize model
rather than general model. For stepwise linear regression analysis of investigating relevant bio
signals to IAQ Evaluation, stress level and heart rate are two more relevant bio-signals to IAQ
Evaluation, moreover, stress level is the most relevant bio-signal to IAQ Evaluation in the
female group, and heart rate is the most bio-signal to IAQ Evaluation in the male group. For
random forest analysis on this investigation, EDA is more relevant bio signal to IAQ Evaluation
for most participants.
Based on the results of stepwise linear regression analysis of investigating relevant parameters
to IAQ Evaluation, indoor relative humidity is the most relevant factor to IAQ Evaluation
among eight factors, and stress level is the most relevant bio-signal to IAQ Evaluation among
four bio-signals. For results of random forest analysis, indoor CO 2 concentration is the most
important parameters among eight parameters, and EDA is most relevant bio-signal to IAQ
Evaluation.
Based on the results of stepwise linear regression of investigating relevant parameters to
thermal comfort. Indoor temperature is the most relevant factor (from indoor temperature,
indoor relative humidity, heart rate, stress level, EDA and skin temperature) to thermal comfort,
and skin temperature is the most relevant bio-signal to thermal comfort. For results of random
forest of investigating relevant parameters to thermal comfort, indoor relative humidity is most
parameter to thermal comfort, and skin temperature is important bio-signal to thermal comfort.

102

Besides, for the accuracy prediction model, in most cases, the accuracy using first day’s data
to train a model and using last day’s data to test the model is higher than the accuracy using
cross validation.
Furthermore, this study showed that half of the participants’ stress level were positively
associated with indoor PM2.5 concentration and indoor CO2 concentration, and half of the
participants’ stress level were negatively correlated with indoor PM 2.5 and indoor CO2
concentration. Therefore, our study did not prove whether the correlation between stress level
and two IAQ parameters (indoor PM2.5 concentration and indoor CO2 concentration) was
positive or negative.
Previous studies have consistently found that high concentrations of PM2.5 can cause lower
HRV (Vallejo et al., 2006; Creason et al., 2001). However, half of the participants’ stress level
in this study positively correlate to indoor PM 2.5 concentration. Furthermore, Kajtár and
Herczeg’s study proved that increased CO2 concentration increased heart rate, but our study
showed that most participants’ heart rate decreases as indoor CO2 concentration rises. In
addition, previous studies found that subjects’ fear and anxiety responses will increase as
indoor CO2 concentration increases. Nevertheless, half of the participants’ stress level in our
study decrease as indoor CO2 concentration increases.
Furthermore, except for the results from the female group based on Pearson's analysis, showing
indoor PM2.5 concentration had a more significant impact on skin temperature and indoor CO2
concentration had a greater impact on EDA and skin temperature, all the other results
(including the Pearson correlation coefficient analysis results of the whole group and male
group, and stepwise linear regression analysis results of the whole group, female group and
male group) about the correlation between human bio-signals and two IAQ parameters showed
that indoor PM2.5 concentration and indoor CO2 concentration had the most significant impact
on heart rate, which proved heart rate was the most relevant bio-signal to indoor CO2
concentration and PM2.5 concentration (Fig 5.1, Fig. 5.5, Fig 5.9, Fig 5.13), and it was also a
more significant bio-signal to IAQ Evaluation (Fig 5.17, Fig 5.22).
6.2 Limitation and future work
6.2.1 short term problems
In this study, every subject had to wear two smartwatches for conductive five days when they
were in their bedroom, even when they were sleeping; some subjects felt a little uncomfortable
sleeping with watches on both hands. In addition, sometimes, when they left their bedroom and
went to the other rooms (such as the living room and bathroom), they tended to forget to take

103

off the watch. Therefore, the physiological responses data measured by the two smartwatches
were the responses in other rooms, which will slightly affect the accuracy of the experimental
results because the sensors for measuring indoor environmental parameters (indoor
temperature, indoor relative humidity, indoor CO2 concentration, and indoor PM2.5
concentration) are all in the experimenter's bedroom.
Besides, the Pa- II- SD sensor detecting PM 2.5 needs to be plugged into a power supply for
normal use, and the sixth experimenter in this experiment accidentally unplugged the power
supply of the PM2.5 sensor, resulting in the loss of three days of indoor PM2.5 data. Moreover,
something went wrong with the Dr.meter LX1330B Digital Illuminance Light Meter measuring
indoor acoustic during its use, resulting in the loss of acoustic data of the last two participants.
Hence, in the future study, two sensors should be placed together and used simultaneously in
case one fails and the data is lost.
6.2.2 long term problems
There are only 12 participants in this study, and the subjects are between 25 and 35 years old,
so the sample size and sample diversity are insufficient. Future studies will need more data
from more participants and more people of various ages. In addition, the experiment period of
each experimenter in this study was only five days, and the whole experiment lasted from
November 2022 to January 2022. Future studies should extend the experiment period of each
experimenter. Each researcher can be asked to conduct experiments for five days a month, so
that the experiment cycle will cover every season and improve the accuracy of research results.
In addition, this study did only focus on residential buildings; future studies could focus on
more residential buildings and commercial buildings.
6.2.3 Additional future work
Based on the results from this study, our future study will focus on developing the IAQ control
model controlling the window to provide natural ventilation to control the IAQ based on the
relevant bio-signals, and then operate, and apply the model in an actual building. Furthermore,
this study applied correlation, stepwise linear regression, and random forest to analyze the data.
Hence, more models could be applied to analyze the data collected in future studies.

104

REFERENCES
Agarwal, N., Meena, C. S., Raj, B. P., Saini, L., Kumar, A., Gopalakrishnan, N., Kumar, A.,
Balam, N. B., Alam, T., Kapoor, N. R., & Aggarwal, V. (2021). Indoor air quality
improvement in COVID-19 pandemic: Review. Sustainable Cities and Society, 70,
102942. https://doi.org/10.1016/J.SCS.2021.102942
Al-Shargie, F. M., Tang, T. B., Badruddin, N., & Kiguchi, M. (2015). Mental Stress
Quantification Using EEG Signals. IFMBE Proceedings, 56, 15–19.
https://doi.org/10.1007/978-981-10-0266-3_4
Al horr, Y., Arif, M., Katafygiotou, M., Mazroei, A., Kaushik, A., & Elsarrag, E. (2016).
Impact of indoor environmental quality on occupant well-being and comfort: A review
of the literature. International Journal of Sustainable Built Environment, 5(1), 1–11.
https://doi.org/10.1016/J.IJSBE.2016.03.006
Allen, J. G., MacNaughton, P., Satish, U., Santanam, S., Vallarino, J., & Spengler, J. D.
(2016). Associations of cognitive function scores with carbon dioxide, ventilation, and
volatile organic compound exposures in office workers: A controlled exposure study of
green and conventional office environments. Environmental Health Perspectives,
124(6), 805–812. https://doi.org/10.1289/EHP.1510037
Bailey, J. E., Argyropoulos, S. V., Kendrick, A. H., & Nutt, D. J. (2005). Behavioral and
cardiovascular effects of 7.5% CO2 in human volunteers. Depression and Anxiety,
21(1), 18–25. https://doi.org/10.1002/DA.20048
Bhagat, R. K., Davies Wykes, M. S., Dalziel, S. B., & Linden, P. F. (2020). Effects of
ventilation on the indoor spread of COVID-19. Journal of Fluid Mechanics, 903.
https://doi.org/10.1017/JFM.2020.720
Bluyssen, P. M. (2013). The healthy indoor environment: How to assess occupants’
wellbeing in buildings. In The Healthy Indoor Environment: How to Assess Occupants’
Wellbeing in Buildings (Vol. 9781315887). https://doi.org/10.4324/9781315887296
Bonato, P. (2005). Advances in wearable technology and applications in physical medicine
and rehabilitation. Journal of NeuroEngineering and Rehabilitation, 2(1), 1–4.
https://doi.org/10.1186/1743-0003-2-2/METRICS
Calì, D., Osterhage, T., Streblow, R., & Müller, D. (2016). Energy performance gap in
refurbished German dwellings: Lesson learned from a field test. Energy and Buildings,
127, 1146–1158. https://doi.org/10.1016/J.ENBUILD.2016.05.020
Centers for Disease Control and Prevention. (n.d.). Retrieved March 4, 2022, from
https://www.cdc.gov/
Chai, Q., Wang, H., Zhai, Y., & Yang, L. (2020). Using machine learning algorithms to

105

predict occupants’ thermal comfort in naturally ventilated residential buildings. Energy
and Buildings, 217, 109937. https://doi.org/10.1016/J.ENBUILD.2020.109937
Chang, C., & Gershwin, M. E. (2004). Indoor air quality and human health. Clinical Reviews
in Allergy & Immunology 2004 27:3, 27(3), 219–239.
https://doi.org/10.1385/CRIAI:27:3:219
Chen, Y. C., & Hsiao, T. C. (2015). Physiological Responses to Different
CO2 Levels in Poor Ventilation Room. IFMBE Proceedings,
45, 423–426. https://doi.org/10.1007/978-3-319-11128-5_106
Choi, J. H., & Loftness, V. (2012). Investigation of human body skin temperatures as a bio-
signal to indicate overall thermal sensations. Building and Environment, 58, 258–269.
https://doi.org/10.1016/J.BUILDENV.2012.07.003
Comunian, S., Dongo, D., Milani, C., & Palestini, P. (2020). Air Pollution and COVID-19:
The Role of Particulate Matter in the Spread and Increase of COVID-19’s Morbidity and
Mortality. International Journal of Environmental Research and Public Health 2020,
Vol. 17, Page 4487, 17(12), 4487. https://doi.org/10.3390/IJERPH17124487
Creason, J., Neas, L., Walsh, D., Williams, R., Sheldon, L., Liao, D., & Shy, C. (2001).
Particulate matter and heart rate variability among elderly retirees: the Baltimore 1998
PM study. Journal of Exposure Science & Environmental Epidemiology 2001 11:2,
11(2), 116–122. https://doi.org/10.1038/sj.jea.7500154
Domingo, J. L., & Rovira, J. (2020). Effects of air pollutants on the transmission and severity
of respiratory viral infections. Environmental Research, 187, 109650.
https://doi.org/10.1016/J.ENVRES.2020.109650
Electroencephalogram (EEG) | Johns Hopkins Medicine. (n.d.). Retrieved March 4, 2022,
from https://www.hopkinsmedicine.org/health/treatment-tests-and-
therapies/electroencephalogram-eeg
Fanger, P. O. (1970). Thermal comfort. Analysis and applications in environmental
engineering. Thermal Comfort. Analysis and Applications in Environmental
Engineering.
Feng, S., Gao, D., Liao, F., Zhou, F., & Wang, X. (2016). The health effects of ambient
PM2.5 and potential mechanisms. Ecotoxicology and Environmental Safety, 128, 67–74.
https://doi.org/10.1016/J.ECOENV.2016.01.030
Forouzanfar, M. H., Afshin, A., Alexander, L. T., Biryukov, S., Brauer, M., Cercy, K.,
Charlson, F. J., Cohen, A. J., Dandona, L., Estep, K., Ferrari, A. J., Frostad, J. J.,
Fullman, N., Godwin, W. W., Griswold, M., Hay, S. I., Kyu, H. H., Larson, H. J., Lim,
S. S., … Zhu, J. (2016). Global, regional, and national comparative risk assessment of
79 behavioural, environmental and occupational, and metabolic risks or clusters of risks,
1990–2015: a systematic analysis for the Global Burden of Disease Study 2015. The

106

Lancet, 388(10053), 1659–1724. https://doi.org/10.1016/S0140-6736(16)31679-8
Ganesh, G. A., Sinha, S. L., Verma, T. N., & Dewangan, S. K. (2021). Investigation of indoor
environment quality and factors affecting human comfort: A critical review. Building
and Environment, 204, 108146. https://doi.org/10.1016/J.BUILDENV.2021.108146
Gunjal, S. G., & Kamalapurkar, S. (n.d.). Air Pollution Prediction Using Machine Learning
Supervised Learning Approach. Retrieved February 19, 2022, from www.ijstr.org
Heart rate: What is a normal heart rate? (n.d.). Retrieved March 4, 2022, from
https://www.medicalnewstoday.com/articles/235710#normal-resting-heart-rate
Holguín, F., Téllez-Rojo, M. M., Hernández, M., Cortez, M., Chow, J. C., Watson, J. G.,
Mannino, D., & Romieu, I. (2003). Air pollution and heart rate variability among the
elderly in Mexico City. Epidemiology, 14(5), 521–527.
https://doi.org/10.1097/01.EDE.0000081999.15060.AE
Jiang, X. Q., Mei, X. D., & Feng, D. (2016). Air pollution and chronic airway diseases: what
should people know and do? Journal of Thoracic Disease, 8(1), E31.
https://doi.org/10.3978/J.ISSN.2072-1439.2015.11.50
Kajtár, L., & Herczeg, L. (n.d.). Influence of carbon-dioxide concentration on human well-
being and intensity of mental work. IDŐJÁRÁS Quarterly Journal of the Hungarian
Meteorological Service, 116(2), 145–169.
Liu, G., Liu, X., Qin, Z., Gu, Z., Wang, G., Shi, W., Wen, D., Yu, L., Luo, Y., & Xiao, H.
(2015). Cardiovascular system response to carbon dioxide and exercise in oxygen-
enriched environment at 3800 m. International Journal of Environmental Research and
Public Health, 12(9), 11781–11796. https://doi.org/10.3390/ijerph120911781
Marquès, M., & Domingo, J. L. (2022). Positive association between outdoor air pollution
and the incidence and severity of COVID-19. A review of the recent scientific
evidences. Environmental Research, 203, 111930.
https://doi.org/10.1016/J.ENVRES.2021.111930
Morawska, L., Tang, J. W., Bahnfleth, W., Bluyssen, P. M., Boerstra, A., Buonanno, G., Cao,
J., Dancer, S., Floto, A., Franchimon, F., Haworth, C., Hogeling, J., Isaxon, C., Jimenez,
J. L., Kurnitski, J., Li, Y., Loomans, M., Marks, G., Marr, L. C., … Yao, M. (2020).
How can airborne transmission of COVID-19 indoors be minimised? Environment
International, 142, 105832. https://doi.org/10.1016/J.ENVINT.2020.105832
Mujeebu, M. A. (2019). Indoor Environmental Quality.
Okamoto-Mizuno, K., & Mizuno, K. (2012). Effects of thermal environment on sleep and
circadian rhythm. Journal of Physiological Anthropology, 31(1), 1–9.
https://doi.org/10.1186/1880-6805-31-14/FIGURES/2

107

Ouedraogo, I., Defourny, P., & Vanclooster, M. (2019). Application of random forest
regression and comparison of its performance to multiple linear regression in modeling
groundwater nitrate concentration at the African continent scale. Hydrogeology Journal,
27(3), 1081–1098. https://doi.org/10.1007/S10040-018-1900-5/TABLES/5
Pandey, P., Patel, D. K., Khan, A. H., Barman, S. C., Murthy, R. C., & Kisku, G. C. (2013).
Temporal distribution of fine particulates (PM2.5, PM10), potentially toxic metals,
PAHs and Metal-bound carcinogenic risk in the population of Lucknow City, India.
Http://Dx.Doi.Org/10.1080/10934529.2013.744613, 48(7), 730–745.
https://doi.org/10.1080/10934529.2013.744613
Paoin, K., Ueda, K., Seposo, X. T., Hayano, J., Kiyono, K., Ueda, N., Kawamura, T., Honda,
A., & Takano, H. (2020). Association between PM2.5 exposure and heart rate variability
for the patients with cardiac problems in Japan. Air Quality, Atmosphere and Health,
13(3), 339–347. https://doi.org/10.1007/S11869-020-00797-8
Park, H., Park, S., & Seo, J. (2020). Evaluation on Air Purifier’s Performance in Reducing
the Concentration of Fine Particulate Matter for Occupants according to its Operation
Methods. International Journal of Environmental Research and Public Health 2020,
Vol. 17, Page 5561, 17(15), 5561. https://doi.org/10.3390/IJERPH17155561
Poirier, B., Guyot, G., Woloszyn, M., Geoffroy, H., Ondarts, M., & Gonze, E. (2021).
Development of an assessment methodology for IAQ ventilation performance in
residential buildings: An investigation of relevant performance indicators. Journal of
Building Engineering, 43, 103140. https://doi.org/10.1016/J.JOBE.2021.103140
Pozzi, R., De Berardis, B., Paoletti, L., & Guastadisegni, C. (2003). Inflammatory mediators
induced by coarse (PM2.5-10) and fine (PM2.5) urban air particles in RAW 264.7 cells.
Toxicology, 183(1–3), 243–254. https://doi.org/10.1016/S0300-483X(02)00545-0
Rashad, M., Khordehgah, N., Żabnieńska-Góra, A., Ahmad, L., & Jouhara, H. (2021). The
utilisation of useful ambient energy in residential dwellings to improve thermal comfort
and reduce energy consumption. International Journal of Thermofluids, 9, 100059.
https://doi.org/10.1016/J.IJFT.2020.100059
Romanovsky, A. A., Romanov-Sky, A. A., & Joseph’, S. (2014). Skin temperature: its role in
thermoregulation. Acta Physiologica, 210(3), 498–507.
https://doi.org/10.1111/APHA.12231
Rudnick, S. N., & Milton, D. K. (2003). Risk of indoor airborne infection transmission
estimated from carbon dioxide concentration. Indoor Air, 13(3), 237–245.
https://doi.org/10.1034/J.1600-0668.2003.00189.X
Sarchiapone, M., Gramaglia, C., Iosue, M., Carli, V., Mandelli, L., Serretti, A., Marangon,
D., & Zeppegno, P. (2018). The association between electrodermal activity (EDA),
depression and suicidal behaviour: A systematic review and narrative synthesis. BMC

108

Psychiatry, 18(1), 1–27. https://doi.org/10.1186/S12888-017-1551-4/TABLES/2
Seddon, K., Morris, K., Bailey, J., Potokar, J., Rich, A., Wilson, S., Bettica, P., & Nutt, D. J.
(2011). Effects of 7.5% CO2 challenge in generalized anxiety disorder. Journal of
Psychopharmacology, 25(1), 43–51. https://doi.org/10.1177/0269881110364270
Starr, D. (2021). The air investigator. Science, 373(6555), 612–615.
https://doi.org/10.1126/SCIENCE.373.6555.612
Suh, H. H., & Zanobetti, A. (2010). Exposure error masks the relationship between traffic-
related air pollution and heart rate variability. Journal of Occupational and
Environmental Medicine, 52(7), 685–692.
https://doi.org/10.1097/JOM.0B013E3181E8071F
Sun, Y., & Zhu, N. (2013). Study on Assessment of High Temperature and Humidity in
Working Environment on Human Health. Advanced Materials Research, 610–613, 739–
742. https://doi.org/10.4028/WWW.SCIENTIFIC.NET/AMR.610-613.739
Turner, W. J. N., & Walker, I. S. (2013). Using a ventilation controller to optimise residential
passive ventilation for energy and indoor air quality. Building and Environment, 70, 20–
30. https://doi.org/10.1016/J.BUILDENV.2013.08.004
U.S. Environmental Protection Agency | US EPA. (n.d.). Retrieved March 4, 2022, from
https://www.epa.gov/
Understanding the stress response - Harvard Health. (n.d.). Retrieved March 4, 2022, from
https://www.health.harvard.edu/staying-healthy/understanding-the-stress-response
Vallejo, M., Ruiz, S., Hermosillo, A. G., Borja-Aburto, V. H., & Cárdenas, M. (2006).
Ambient fine particles modify heart rate variability in young healthy adults. Journal of
Exposure Science and Environmental Epidemiology, 16(2), 125–130.
https://doi.org/10.1038/SJ.JEA.7500447
van der Heide, E. M. M., Veerkamp, R. F., van Pelt, M. L., Kamphuis, C., Athanasiadis, I., &
Ducro, B. J. (2019). Comparing regression, naive Bayes, and random forest methods in
the prediction of individual survival to second lactation in Holstein cattle. Journal of
Dairy Science, 102(10), 9409–9421. https://doi.org/10.3168/JDS.2019-16295
Vilcassim, M. J. R., Thurston, G. D., Chen, L. C., Lim, C. C., Saunders, E., Yao, Y., &
Gordon, T. (2019). Exposure to air pollution is associated with adverse cardiopulmonary
health effects in international travellers. Journal of Travel Medicine, 26(5), 1–8.
https://doi.org/10.1093/JTM/TAZ032
Wu, C. F., Kuo, I. C., Su, T. C., Li, Y. R., Lin, L. Y., Chan, C. C., & Hsu, S. C. (2010).
Effects of Personal Exposure to Particulate Matter and Ozone on Arterial Stiffness and
Heart Rate Variability in Healthy Adults. American Journal of Epidemiology, 171(12),
1299–1309. https://doi.org/10.1093/AJE/KWQ060

109

Wu, X., Nethery, R. C., Sabath, B., Braun, D., Dominici, F., & James, C. (2020). Exposure to
air pollution and COVID-19 mortality in the United States: A nationwide cross-sectional
study. Science Advances. https://doi.org/10.1101/2020.04.05.20054502
Wysocka, M. (2018). DALY indicator as an assessment tool for indoor air quality impact on
human health. E3S Web of Conferences, 49, 00133.
https://doi.org/10.1051/E3SCONF/20184900133
Xu, D., Huang, N., Wang, Q., & Liu, H. (2008). [Study of ambient PM2.5 on the influence of
the inflammation injury and the immune function of subchronic exposure rats]. Wei
Sheng Yan Jiu = Journal of Hygiene Research, 37(4), 423–428.
https://europepmc.org/article/med/18839524
Zhang, X., Huettel, S. A., Mullette-Gillman, O. A., Guo, H., & Wang, L. (2019). Exploring
common changes after acute mental stress and acute tryptophan depletion: Resting-state
fMRI studies. Journal of Psychiatric Research, 113, 172–180.
https://doi.org/10.1016/J.JPSYCHIRES.2019.03.025
Zoran, M. A., Savastru, R. S., Savastru, D. M., & Tautan, M. N. (2020). Assessing the
relationship between surface levels of PM2.5 and PM10 particulate matter impact on
COVID-19 in Milan, Italy. The Science of the Total Environment, 738.
https://doi.org/10.1016/J.SCITOTENV.2020.139825

Abstract (if available)

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

Impact of bamboo materials on indoor air quality

PDF

Impacts of indoor environmental quality on occupants environmental comfort: a post occupancy evaluation study

PDF

Using bio-signals with smart windows

PDF

Impacts of building performance on occupants' work productivity: a post occupancy evaluation study

PDF

Enhancing thermal comfort: air temperature control based on human facial skin temperature

PDF

Multi-occupancy environmental control for smart connected communities

PDF

Vegetated facades as environmental control systems: filtering fine particulate matter (PM2.5) for improving indoor air quality

PDF

Exploring participatory sensing and the Internet of things to evaluate temperature setpoint policy and potential of overheating/overcooling of spaces on the USC campus

PDF

Nursing comfort study: are hospitals performing for nurses? IEQ analysis of working conditions inside hospitals

PDF

IEQ, sleep quality, and IoT: meta-analysis on improving IEQ and sleep quality using IoT

PDF

Human–building integration: machine learning–based and occupant eye pupil size–driven lighting control as an applicable visual comfort tool in the office environment

PDF

Exploration for the prediction of thermal comfort & sensation with application of building HVAC automation

PDF

Human-building integration: Investigation of human eye pupil sizes as a measure of visual sensation in the workstation environment

PDF

Human-environmental interaction: potential use of pupil size for office lighting controls

PDF

Developing environmental controls using a data-driven approach for enhancing environmental comfort and energy performance

PDF

Considering occupants: comprehensive POE research on office environment of Southern California

PDF

Human-building integration based on biometric signal analysis: investigation of the relationships between human comfort and IEQ in a multi-occupancy condition

PDF

Personal PM2.5 exposure during pregnancy in an environmental health disparities population

PDF

Single & multi-family housing: the complex behind the simple: a kit of parts for minimally invasive interventions

PDF

Quantify human experience: integrating virtual reality, biometric sensors, and machine learning

Asset Metadata

Creator Gong, Minghuan (author)

Core Title Indoor air quality for human health in residential buildings

School School of Architecture

Degree Master of Building Science

Degree Program Building Science

Degree Conferral Date 2022-08

Publication Date 06/22/2022

Defense Date 04/20/2022

Publisher University of Southern California (original), University of Southern California. Libraries (digital)

Tag human health,human physiological responses,indoor air quality,indoor environmental quality,OAI-PMH Harvest

Language English

Contributor Electronically uploaded by the author (provenance)

Advisor Choi, Joon-Ho (committee chair), Habre, Rima (committee member), Konis, Kyle (committee member), Park, Dong Yoon (committee member)

Creator Email gongming@usc.edu,minghuangong@gmail.com

Permanent Link (DOI) https://doi.org/10.25549/usctheses-oUC111345520

Unique identifier UC111345520

Legacy Identifier etd-GongMinghu-10780

Document Type Thesis

Rights Gong, Minghuan

Type texts

Source 20220623-usctheses-batch949 (batch), University of Southern California (contributing entity), University of Southern California Dissertations and Theses (collection)

Access Conditions The author retains rights to his/her dissertation, thesis or other graduate work according to U.S. copyright law. Electronic access is being provided by the USC Libraries in agreement with the author, as the original true and official version of the work, but does not grant the reader permission to use the work if the desired use is covered by copyright. It is the author, as rights holder, who must provide use permission if such use is covered by copyright. The original signature page accompanying the original submission of the work to the USC Libraries is retained by the USC Libraries and a copy of it may be obtained by authorized requesters contacting the repository e-mail address given.

Repository Name University of Southern California Digital Library

Repository Location USC Digital Library, University of Southern California, University Park Campus MC 2810, 3434 South Grand Avenue, 2nd Floor, Los Angeles, California 90089-2810, USA

Repository Email cisadmin@lib.usc.edu