University of Southern California Dissertations and Theses

Indoor environmental quality and comfort: IEQ adaptation and human physiological responses in commercial buildings

(USC Thesis Other)

Indoor environmental quality and comfort: IEQ adaptation and human physiological responses in commercial buildings

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content

INDOOR ENVIRONMENTAL QUALITY AND COMFORT:

IEQ ADAPTATIONS AND HUMAN PHYSIOLOGICAL RESPONSES IN COMMERCIAL
BUILDINGS

by

HAOYUE DAI

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

May 2023

ii
Acknowledgements
I would like to express my sincere gratitude to all those who have supported me during the completion of
this project. First and foremost, I want to thank my supervisor, Professor Joon-ho Choi, for his guidance,
encouragement, and support throughout this research endeavor. Without his insights and expertise, this
project would not have been possible. I would also like to thank my two committee members, Professor
Gideon Susman and Professor Kristina Lerman, who provided valuable feedback and suggestions, which
helped me refine my ideas and strengthen my arguments. Your contributions were instrumental in shaping
the final version of this project. In addition, I am grateful to my classmates and colleges for being my
volunteers with unwavering support and encouragement during this challenging time. Your love and
support gave me the motivation and strength to persevere through the difficulties of this project.
Once again, thank you to all those who have supported me throughout this project. Your assistance and
encouragement have been invaluable, and I am deeply grateful for all that you have done.

iii
TABLE OF CONTENTS
Acknowledgements ....................................................................................................................................... ii
List of Tables ............................................................................................................................................... vi
List of Figures ............................................................................................................................................. vii
Abstract ........................................................................................................................................................ ix
CHAPTER 1: INTRODUCTION ................................................................................................................. 1
1.1 “Environmental and Social” in ESG .............................................................................................. 2
1.2 The impact of Indoor Environmental Quality on human comfort ...................................................... 4
1.2.1 The significance of Indoor Air Quality in the building environment ..................................... 5
1.2.2 The significance of Lighting on human lives ......................................................................... 7
1.2.3 The significance of Thermal conditions in human satisfaction .............................................. 7
1.2.4 The significance of Acoustics on human health ..................................................................... 8
1.3 Human physiological responses by using biometrics technology ................................................. 9
1.3.1 The connection of Skin temperature, Electrical activity of skin with human emotions ......... 9
1.3.2 Heart rate with human health ............................................................................................... 10
1.3.3 Electroencephalography (EEG) with human brain .............................................................. 11
1.4 The application and impact of Ventilation ........................................................................................ 12
1.4 Summary ........................................................................................................................................... 13
CHAPTER 2: BACKGROUND AND LITERATURE REVIEW ............................................................. 15
2.1 The impact of indoor environmental quality ..................................................................................... 15
2.2 The assessment of indoor environmental quality indicators ............................................................. 17
2.2.1 Indicators of Indoor Air Quality current technologies ............................................................... 18
2.2.2 Indicators of Thermal Comfort existing model .......................................................................... 20
2.2.3 Indicators of Lighting and Acoustics ......................................................................................... 22
2.3 The correlation between Indoor environmental quality and human physiological responses........... 23
2.4 Current technology strategies in smart ventilation control for energy efficiency ............................. 25
2.4.1 Benefits and limitation of smart ventilation system ................................................................... 26
2.4.2 Applications of smart ventilation ............................................................................................... 28
2.5 Data analyses approaches in human physiological responses and IEQ domain ............................... 29
2.6 Summary ........................................................................................................................................... 31

iv
CHAPTER 3: METHODOLOGY .............................................................................................................. 32
3.1 Project scope ..................................................................................................................................... 32
3.2 Experimental procedures ................................................................................................................... 33
3.3 Data collected .................................................................................................................................... 38
3.3.1 Data for Indoor Environmental Quality ..................................................................................... 38
3.3.2 Data for Surveys ......................................................................................................................... 43
3.3.3 Data for participants’ physiological responses ........................................................................... 47
3.4 Data analysis approach: Cross correlation analysis and Machine Learning ..................................... 49
3.5 Summary ........................................................................................................................................... 50
CHAPTER 4: DATA ANALYSIS ............................................................................................................. 51
4.1 Data organization .............................................................................................................................. 51
4.2 Cross Correlation analysis ................................................................................................................ 54
4.2.1 Applied analysis approaches ...................................................................................................... 54
4.2.2 Processing the developed Cross correlations by person ............................................................. 57
4.2.3 Processing the developed Cross correlations by gathering all participants’ data ....................... 63
4.3 The impact of IEQ factors on Human Physiological responses indicators ....................................... 64
4.3.1 Ranking significance level by person ......................................................................................... 64
4.3.2 Ranking significance level by genders ....................................................................................... 67
4.3.3 Ranking significance level by experimental spaces ................................................................... 71
4.3.4 Ranking significance level by gathering all participants’ data ................................................... 74
4.4 Summary ........................................................................................................................................... 76
CHAPTER 5: DISCUSSION OF THE PREDICTION MODELLING ..................................................... 78
5.1 Theory for establishing the prediction models .................................................................................. 78
5.1.1 Supervised learning .................................................................................................................... 78
5.1.2 Random Forest ........................................................................................................................... 79
5.2 Prediction model results and accuracy .............................................................................................. 81
5.2.1 Individual prediction models ...................................................................................................... 81
5.2.2 General prediction model ........................................................................................................... 86
5.3 Comparison and discussion of the results ......................................................................................... 87
5.3.1 Comparison between the individual prediction model and general prediction model ............... 87
5.3.2 Discussion of participants’ weight, height, and BMI with individual prediction ....................... 90
5.4 Summary ........................................................................................................................................... 93
CHAPTER 6: CONCLUSIONS AND FUTURE WORK .......................................................................... 94
6.1 Conclusion ........................................................................................................................................ 94

v
6.2 Future Work ...................................................................................................................................... 96
6.2.1 Limitation of the current workflow ............................................................................................ 96
6.2.2 Recommendations for future practice ........................................................................................ 98
BIBLIOGRAPHY ..................................................................................................................................... 103

vi
List of Tables
Table 3.1 Experiment devices ..................................................................................................................... 35
Table 3.2 HOBO MX1102 Details ............................................................................................................. 39
Table 3.3 Pa- II- SD Details ........................................................................................................................ 41
Table 3.4 Dr.Meter LX1330B Digital Illuminance Light Meter Details .................................................... 42
Table 3.5 PCE-SDL 1 Details ..................................................................................................................... 43
Table 3.6 Garmin Connect Details .............................................................................................................. 47
Table 3.7 Empatica embrace 2 Details ........................................................................................................ 48
Table 4.1 p values results of participant 1 .................................................................................................. 61
Table 4.2 p values results removing p>0.05 of participant 1 ...................................................................... 61
Table 4.3 p values final results of participant 1 .......................................................................................... 61
Table 4.4 Combined results from TLCC and p values of participant 1 ...................................................... 62
Table 4.5 Final selected correlations for forecasting of participant 1 ......................................................... 63
Table 4.6 p values final results based on combining all the data ................................................................ 63
Table 4.7 Combined results from TLCC and p values based on combining all the data ............................ 64
Table 4.8 Final selected correlations for forecasting based on combining all the data ............................... 64
Table 4.9 The number of significant correlation cases by persons ............................................................. 65
Table 4.10 The number of significant correlation cases by gender - Females ............................................ 68
Table 4.11 The number of significant correlation cases by gender – Males ............................................... 68
Table 4.12 The number of significant correlation cases by spaces – DT office ......................................... 71
Table 4.13 The number of significant correlation cases by spaces – USC Studio ...................................... 71
Table 4.14 The significant correlation cases p values based on all the combined individual datasets ....... 75
Table 5.1 Participants' BMI organized according to IAQ individual prediction accuracy ......................... 90
Table 5.2 Participants' BMI organized according to Thermal Comfort individual prediction accuracy .... 91

vii
List of Figures
Figure 1.1 ESG concept ································ ································ ······························· 3
Figure 1.2 The impact of IAQ ································ ································ ························ 6
Figure 1.3 Some examples of wearable devices ································ ································ ··· 9
Figure 1.4 Heart rate in Apple Watch ································ ································ ·············· 11
Figure 2.1 IAQ indicators and measurement ································ ································ ······ 20
Figure 2.2 The main factors impacting thermal comfort ································ ························· 21
Figure 2.3 A seven-point scale of PMV ································ ································ ············ 22
Figure 2.4 Main features of smart ventilation and responsible parameters for smart ventilation system ·· 26
Figure 2.5 Personalized ventilation in use ································ ································ ········· 29
Figure 3.1 Whole progress ································ ································ ··························· 32
Figure 3.2 DT office area ································ ································ ···························· 33
Figure 3.4 Methodology Diagram ································ ································ ·················· 37
Figure 3.4 HOBO in experiment ································ ································ ···················· 40
Figure 3.5 Pa-II-SD in experiment································ ································ ·················· 41
Figure 3.6 Dr.Meter LX1330B Digital Illuminance Light Meter in experiment ······························ 42
Figure 3.7 PCE-SDL 1 in the experiment ································ ································ ·········· 43
Figure 3.8 Indoor Air Quality Evaluation survey ································ ································ · 45
Figure 3.9 Indoor Thermal Comfort Evaluation Survey ································ ························· 46
Figure 3.10 Smartwatches in the experiment ································ ································ ······ 49
Figure 4.1 Example of stress level and heart rate data ································ ··························· 52
Figure 4.2 Example of EDA and Skin temperature ································ ······························· 52
Figure 4.3 Example of organized dataset ································ ································ ·········· 54
Figure 4.4 Cross correlation function for TLCC analysis ································ ························ 55
Figure 4.5 WTLCC coding ································ ································ ·························· 56
Figure 4.6 Granger Causality Test coding ································ ································ ········· 57
Figure 4.7 WTLCC - EDA vs Temperature – Participant 1 ································ ····················· 58
Figure 4.8 TLCC - EDA vs Temperature – Participant 1 ································ ························ 58
Figure 4.9 WTLCC – Heart rate vs Temperature – Participant 1 ································ ··············· 59
Figure 4.10 TLCC – Heart rate vs Temperature – Participant 1 ································ ················ 60
Figure 4.11 Heatmap showing the percentage of significant correlations cases based on personal data ··· 65
Figure 4.12 Average coefficient heatmap based on personal data ································ ·············· 66
Figure 4.13 Histogram for the significant correlation cases for each bio-signal based on personal data ·· 66

viii
Figure 4.14 Histogram for the significant correlation cases for IEQ indicators based on personal data ··· 67
Figure 4.15 Heatmap showing the percentage of significant correlations cases by gender - Males ········ 68
Figure 4.16 Heatmap showing the percentage of significant correlation cases by gender - Females ······· 69
Figure 4.17 The comparison between females and males for each bio-signals ······························· 70
Figure 4.18 The comparison between females and males for each IEQ indicator ···························· 70
Figure 4.19 Heatmap showing the significance of correlations summarized by spaces -DT office ········ 72
Figure 4.20 Heatmap showing the significance of correlations summarized by spaces – USC studio ····· 73
Figure 4.21 Histogram for the significant correlation cases of bio-signals based on USC studio data ····· 73
Figure 4.22 Histogram for the significant correlation cases of IEQ indicators based on USC studio data 74
Figure 4.23 Heatmap showing the significant correlation cases based on combined individual datasets ·· 75
Figure 4.24 Average coefficient heatmap based on all the combined individual datasets ··················· 76
Figure 5.1 Random Forest prediction model algorithm example ································ ··············· 80
Figure 5.2 The difference between IAQ evaluation prediction data and actual data-Participant 1 ········· 82
Figure 5.3 The difference between Thermal Comfort evaluation prediction and actual data-Participant 1 82
Figure 5.4 The accuracies of IAQ evaluation prediction models ································ ··············· 83
Figure 5.5 The accuracies of Thermal Comfort evaluation prediction models ································ 84
Figure 5.6 T-test for IAQ evaluation prediction accuracies by genders(p=0.146) ···························· 85
Figure 5.6 T-test for Thermal Comfort evaluation prediction accuracies by genders (p=0.473) ············ 85
Figure 5.7 IAQ evaluation prediction accuracies comparison ································ ··················· 86
Figure 5.8 Thermal Comfort evaluation prediction accuracies comparison ································ ··· 87
Figure 5.9 All IAQ evaluation prediction accuracies comparison ································ ·············· 88
Figure 5.10 All Thermal Comfort evaluation prediction accuracies comparison ····························· 89
Figure 5.11 Participants’ BMI and average BMI according to IAQ prediction accuracy rank ·············· 92
Figure 5.12 Participants’ BMI and average BMI according to Thermal Comfort accuracy rank ··········· 92
Figure 6.1 Single-patient room ································ ································ ······················ 99
Figure 6.2 Predicted HVAC system in suggested Kindergarten environment ······························· 100
Figure 6.3 Predicted HVAC system in the suggested meeting room ································ ········· 101

ix
Abstract:
During the COVID-19 pandemic, Indoor Environmental Quality (IEQ) was considered more seriously by
the public because it pertained to the risk of infection. Especially for commercial buildings, the indoor
environment was linked with occupants’ health and productivity as well as the company’s performance
according to ESG criteria. However, even though there were many studies evaluated IEQ, the experiments
were not focused on developing a predictive model between human physiological responses and indoor
environmental quality. This research investigated the correlations between human physiological responses
and IEQ components in two office areas with 14 participants. The IEQ data, including indoor air quality,
thermal comfort, lighting, and acoustics, were collected using sensors, while simultaneously wearable
devices were used to record human physiological response data. Cross-correlation analysis was applied to
establish the correlations between indicators when analyzing datasets from sensors and wearable devices.
Then predicted models were generated for personal and general data by using supervised machine learning
for IAQ and thermal evaluation based on human bio-signals. As a result, this research confirmed that the
human physiological signals provided feasibility to predict the user’s environmental satisfaction as a
function of the data-driven model at both the individual and the general models based on the data of multiple
human subject experiments. Even though this study provided a robust conclusion based on the analyzed
results of the dataset, the prediction performance could be better by accommodating more datasets with
more participants from different ages and occupied environments for a future study.

Keywords: Indoor environmental quality, Thermal comfort, Bio-signals, Human comfort

x
Hypothesis:
1. Human physiology response relevant to the IEQ indicators and bio-signals can be applied to
predict environmental satisfaction in the form of an individual model and a general one,
respectively.
2. Furthermore, the accuracy of the prediction model generated by personal data is higher than
the other based on the general data.

Aims and Objectives:
1. Identify the correlation between indoor air quality and human physiological responses.
2. Generate and validate the predictive model for indoor air quality based on human physiology responses.
3. Compare the predictive model accuracy based on individual data with the predictive model accuracy
based on dataset combining all data.

1
CHAPTER 1: INTRODUCTION
In commercial buildings, Indoor Environmental Quality (IEQ) has become increasingly linked with a
company’s profitability and sustainability. One of the reasons for this is the popularity of Environmental,
Social, and Governance (ESG) criteria. As one of the criteria in ESG, “E (Environmental)” emphasizes the
large impact of the environment on investors, stakeholders, employees, and customers who are occupants
of office buildings (What Is Environmental, Social, and Governance (ESG) Investing?, n.d.). In other words,
the E concept shows that occupants’ responses are related to Indoor Environmental Quality in commercial
buildings. However, human responses are varied because of different thermal histories and living habits.
This characteristic makes the evaluation of physiological responses hard. Therefore, there are rare studies
investigating the correlation between IEQ indicators and human response indicators, especially in
commercial buildings. In order to improve Indoor Environmental Quality, this research focuses on
establishing the correlations between these indicators to predict IEQ conditions by using human
physiological responses. Additionally, researchers in the previous work have already investigated the
correlations between human physiological responses and IEQ indicators in small-family houses and
apartments. This research applies the previous methodology to a larger office building so as to not only
validate the previous outcome but also improve the testing scales.
This chapter starts with the “E” concept in ESG, and then introduces Indoor Environmental Quality and its
components. Additionally, human physiological responses are presented including indicators and biological
mechanisms. Finally, the impacts of ventilation, as one of the most effective ways to improve IEQ, are

2
described.
1.1 “Environmental and Social” in ESG
As a set of standards, Environmental, Social, and Governance (ESG) criteria are popularly used by
stakeholders to understand a company’s behavior in managing risks and opportunities (What Is
Environmental, Social, and Governance (ESG) Investing?, n.d.). This term describes the sustainable
operations of an organization so that it is often concerned by not only investors, and stakeholders, but also
customers, suppliers, and employees. In detail, the “E” stands for Environmental criteria which evaluate
environmental impacts and risk management for an organization(What Is ESG (Environmental, Social &
Governance)?, n.d.). It includes Climate Change, Natural Resources, Pollution & Waste, and
Environmental Opportunities (What Is Environmental, Social, and Governance (ESG) Investing?, n.d.); the
letter “S” is the Social criteria for a company’s relationship with stakeholders, such as Human Capital
Management metrics; the letter “G” refers to Governance describing leadership and management. It
expresses accountability to stakeholders through accurate and transparent accounting methods, leadership
integrity, and diversity(What Is ESG (Environmental, Social & Governance)?, n.d.).
Environmental factors can be divided into four parts: climate change, natural resources, pollution and waste,
and environmental opportunities(Understanding ESG: Environmental Factors | M&H Engineering
& Consulting for The Oil & Gas Industry, n.d.). Regarding climate change, companies should reduce
greenhouse gases emission, and improve carbon efficiency. When it comes to natural resources, it focuses
on water conservation, biodiversity protection, and responsible land use. In terms of pollution and waste,
the organizations are responsible for reducing solid waste, toxic emissions, water pollution, and electronic

3
waste. Furthermore, environmental opportunities are recommended to apply, such as clean technology and
green building practice(Understanding ESG: Environmental Factors | M&H Engineering &
Consulting for The Oil & Gas Industry, n.d.).
Compared with the other two factors, environmental criteria is the most visible and high-profile component
of ESG(Understanding ESG: Environmental Factors | M&H Engineering & Consulting for The Oil
& Gas Industry, n.d.). For investors, customers and stakeholders, environmental factors can be seen inside
and outside of physical buildings and surrounding sites, so this characteristic makes E more transparent and
real to revealing environmental conditions in an organization. Additionally, nowadays, climate change,
water problems, and carbon footprint are getting more and more attention globally so the building
environment becomes the direct criteria for evaluating the financial impact and competitive position of a
company(Understanding ESG: Environmental Factors | M&H Engineering & Consulting for The
Oil & Gas Industry, n.d.). Therefore, building an environment is important for companies to enhance their
potential values.

Figure 1.1 ESG concept (WONGTRAKOOL, 2018)

4
From figure 1.1 ESG concept, there are many branches belonging to Environmental, Social, and
Governance. All of the branches are related to each other showing cooperation in evaluation. There are two
branches under Social criteria that have close relationships with Environmental criteria: one is Working
conditions and another one is Health and safety. Working conditions decide occupants’ environmental
satisfaction, while Health and safety are attributed to people’s well-being and physiological benefits. In the
current community, especially in the building industry, quantifying environmental and physiological
benefits is a key aspect when estimating and implementing any relevant design, construction, and operation
strategies because one of the large expanses of cost is personnel salaries and benefits. If companies can
quantify environmental and physiological benefits, it is possible to reduce extra operational costs in the
design, and construction stages and improve profit through environmental and physiological benefits.
Therefore, it is significant to develop approaches to quantifying environmental and physiological benefits
to enhance human environmental well-being and satisfaction.

1.2 The impact of Indoor Environmental Quality on human comfort
Indoor Environmental Quality (IEQ) is the general term to describe the interior condition of the building,
including air quality, lighting, thermal conditions, and acoustics (LEED, 2014). IEQ has many effects on
human health and productivity. Good IEQ is effective in lowering the risks of human illness and providing
high productivity (LEED, 2014), while unsatisfied IEQ leads to Sick Building Syndrome (SBS). SBS
describes occupants’ unhealthy or uncomfortable symptoms which are directly linked to building
experience(Joshi, 2008). The common SBS include headache, dizziness, nausea, eye, nose or throat

5
irritation, dry cough, dry or itching skin, difficulty in concentration, fatigue, sensitivity to odors, hoarseness
of voice, allergies, cold, flu-like symptoms, increased incidence of asthma attacks and personality
changes(Joshi, 2008). Therefore, it is significant to pay attention to the impact of Indoor Environment
Quality for enhancing human comfort and satisfaction. For a company, the strategy of employees’ health
and productivity is the key point of the company’s operating cost since one of the large expenses is the
personnel salaries and benefits. Therefore, for commercial buildings, considering the company’s cost and
performance, it is necessary for companies to focus on enhancing IEQ. There are 4 aspects used to quantify
the building environment that will be introduced in the following part: air quality, lighting, and thermal
condition.
1.2.1 The significance of Indoor Air Quality in the building environment
According to the EPA (Environmental Protection Agency), Indoor air quality (IAQ) refers to the air quality
inside or around buildings which relates to occupants’ health and comfort. IAQ is defined by the
concentrations of chemical, microbiological contamination, and physical factors (Guillard, 2021). Some of
these pollutants come from building materials, furnishings, and products. Others are related to indoor
activities, such as smoking and redecorating (Indoor Air and Coronavirus (COVID-19) | US EPA, n.d.). In
order to assess indoor air quality, a series of parameters for measuring various sources of pollution should
be taken into account. These parameters mainly include Carbon Dioxide (CO₂), relative humidity, indoor
temperature, the famous Volatile Organic Compounds (VOCs: cleaning products, disinfectants, paints,
coatings, etc.), PM2.5 and PM10 particles (dust in suspension) (Guillard, 2021). According to the news,
people spend almost 94% of their time inside buildings during work and entertainment (Guillard, 2021).

6
However, this long-time indoor activity can lead to air indoor pollution contacting the skin and being
absorbed in the respiratory system. This causes some immediate effects and long-term effects on human
health, and it is known as Sick Building Syndrome. For short-term issues, exposures to air pollution can
irritate many organs, such as the eyes and nose, and result in headaches, dizziness, and fatigue (Indoor Air
and Coronavirus (COVID-19) | US EPA, n.d.). Additionally, some research shows that 75% of COVID-19
infections happen in enclosed spaces (Guillard, 2021). When it comes to long-term effects, IAQ affects
physical health as well as mental health. Respiratory, heart and cardiovascular diseases are the common
effects caused by indoor pollution. These physical illnesses are severely debilitating and fatal to the human
body and health (Indoor Air and Coronavirus (COVID-19) | US EPA, n.d.). Regarding with mental health,
some symptoms of behaviors occur on people, such as lack of concentration, reduced productivity,
increased absenteeism, and even stress and depression.
Therefore, considering these harmful influences of bad Indoor Air Quality, it is significant to maintain a
good IAQ to avoid instantaneous and permanent impact to human health.

Figure 1.2 The impact of IAQ (Air Control Guy, n.d.)

7
1.2.2 The significance of Lighting on human lives
Lighting, as one of the components of IEQ, offers a wealth of benefits for human health and life. Adequate
lighting provides comfortable visible performance and safety which can avoid injuries and falls.
Additionally, lighting is essential for human health and well-being because the regulation of bodily function
is based on light, especially for the function of the nervous and endocrine systems (Osibona et al., 2021).
For example, the releasing of melatonin is based on how much light is received to regulate the body’s
circadian rhythm; the Hormone is also controlled by light to adjust sleep and alertness(Osibona et al., 2021).
Therefore, if there is improper lighting exposure, these rhythms will be disrupted resulting in an unhealthy
body.
There are two categories of indoor lighting: daylighting and artificial lighting. Natural light can illuminate
buildings through windows in various directions according to the time(Osibona et al., 2021). Artificial light
can be supplied with natural light in the day without too much sun and continue at night to guarantee indoor
activities. Combining these two kinds of lighting can generate a comfortable indoor lighting condition for
human lives. Therefore, lighting also plays a key role in human health.
1.2.3 The significance of Thermal conditions in human satisfaction
Thermal comfort is defined as the person’s psychological state of mind which expresses satisfaction with
the thermal condition (Thermal Comfort in Buildings - Designing Buildings, n.d.). It is difficult to describe
in degrees because the human thermal environment is not straightforward. Also, thermal experience is a
personal description that is dependent on various criteria and is different from one person to another one
even if they are in the same space. There are six basic factors for thermal comfort: for environmental factors,

8
air temperature, radiant temperature, air velocity, and humidity are included; for personal factors, clothing
insulation, and metabolic heat are considered. All these six factors decide how people feel about the
environment. The Health and Safety Executive (HSE) suggested that reasonable comfort can be achieved
when at least 80% of occupants express thermal satisfaction (Thermal Comfort in Buildings - Designing
Buildings, n.d.). Therefore, doing surveys with occupants can assess thermal comfort satisfaction.
Thermal comfort affects human work performance and productivity(Thermal Comfort in Buildings -
Designing Buildings, n.d.). However, it is common for problems with air-conditioning and heating in the
workplace(SafeWork NSW, n.d.). Some old buildings’ outdated HVAC systems lead to uneven distribution
of temperature in one space. Additionally, at the different times of day and seasons of the year, the
temperature cannot be controlled well in different building areas. Therefore, it is significant to improve the
building’s HVAC system to provide satisfied thermal comfort.
1.2.4 The significance of Acoustics on human health
Acoustic comfort defines as an occupant’s sound experience in a building. There are three main noise issues:
noise from outside, adjacent buildings, and in space(NJ Green Building Manual, n.d.). Excessive noise
affects people’s efficiency, mood, stress, and productivity. Numerous research have found that noise leads
to ambient pressure in the workplace(Sound Zero, n.d.). Furthermore, this pressure is relative to job
satisfaction and causes productivity reduction. Long-term noise problems can contribute to illness and
absenteeism or turnover in staff. Therefore, as one of the components of IEQ, acoustic should be concerned
seriously.

9
1.3 Human physiological responses by using biometrics technology
Biometrics technology generally is used to identify a person by biological characteristics (Dastbaz et al.,
2013). This method has already been popular in human security checks, mobile access and authentication,
and health recording by recording human physiological responses which are unique for everyone. These
biometric data can be collected in a centralized database so that authentication or identification can be
completed in the cloud database without direct access to the local biometric data itself (What Is Biometrics?,
n.d.). The wearable device is one of the most common applications of Biometrics. They are usually utilized
as a wristband over the pulse to measure individual biometrics information, such as heart rate, skin
temperature, and calories burned. These parameters give people more opportunities for self-diagnosis as
well as prompt access to treatment when an emergency happens. This technology is used for quantifying
human physiological benefits in this study through indicators from wearable devices, such as skin
conductivity, heart rate, Electroencephalography (EEG) and Electrical activity of skin (EDA).

Figure 1.3 Some examples of wearable devices (Mod, n.d.)
1.3.1 The connection of Skin temperature, Electrical activity of skin with human emotions
Skin temperature is a result of the heat balance between the body's interior and its surroundings. Medically,

10
it is a manifestation of the heat on the outermost surface of the body, which is lower than the temperature
of the body interior. The normal range of skin temperature is from 33º C to 37 º C (WHOOP, n.d.). If the
skin temperature is higher than the usual situation, it indicates fever or other illness, while a lower
temperature means hypothermia which leads the body draws off surface heat to the body inside organs.
Electrical activity of skin (EDA) is also named galvanic skin response. It refers to the variation of the skin
conductance responding to sweat secretion (Farnsworth, 2019). It tested the change of skin conductance
when applying an undetectable voltage to the skin by using EDA devices. Similarly to skin temperature,
EDA is related to internal temperature, but It also shows a strong association with emotion. When people
have positive emotions, such as happy or joyful, or negative emotions, such as threatening and sad, EDA
increases because of these emotional arousals.
1.3.2 Heart rate with human health
Heart rate is measured by the number of times that heart beats in one minute. Heart rate is typically faster
when doing sports or feeling excited or scared. When people are resting or comfortable, the heart rate will
be back to the usual case. This variability is not simply random fluctuations but a traceable response to
internal and external stimuli(Draghici & Taylor, 2016). Additionally, heart rate identifies a range of health
issues. An irregular heartbeat can be a sign of serious health conditions; if there are diseases or injury
weakness of the heart, the organs are possible to receive insufficient blood (MacGill, 2021). Furthermore,
heart rate, as one of the representations of cardiac chronotropy, is accessible by simple palpation of an
artery. That is the reason why heart rate is a popular indicator used in wearable devices.

11

Figure 1.4 Heart rate in Apple Watch (V. Song, 2022)
1.3.3 Electroencephalography (EEG) with human brain
Electroencephalography (EEG) is a recording of brain activity that is utilized to help diagnose and monitor
brain conditions (Electroencephalogram (EEG) - NHS, n.d.). It is measured by using small, metal electrodes
attached to the scalp. Because brain cells continuously communicate via electrical impulses even during
sleep, EEG is a significant indicator for brain diagnosis. EED is recommended by healthcare providers to
diagnose several types of brain disorders. For example, for epilepsy patients, there are rapid spiking waves
on the EEG; Alzheimer's disease and certain psychoses can lead to unusual EEG waves
(Electroencephalogram (EEG) | Johns Hopkins Medicine, n.d.). Additionally, stress can be recognized by
using EEG signal. A certain range of frequencies of EEG is caused by stress: beta waves and alpha waves
are different and the change of their ratio shows the degree of stress encountered (Waili et al., 2020).

12

Figure 1.5 Examples of Electroencephalography (Adamou et al., 2020)
1.4 The application and impact of Ventilation
Ventilation is recognized as an effective approach to controlling and reducing airborne diseases (Qian &
Zheng, 2018). In detail, ventilation can decrease the cross infection by removing or diluting virus-laden
aerosols exhaled by infected patients. The purposes of ventilation include outdoor air supply, excess heat
release, dehumidification, and contaminant movement to provide indoor health and comfort. After
introducing IEQ and human physiological responses, as research targets, the results of this study will
contribute to helping reduce the occupants’ exposure to any potential air pollution by optimizing ventilation
systems. Meanwhile, any potential risk, caused by organic pollutants, such as COVID and SARS, can be
minimized.
One of the examples is the 2003 worldwide SARS. The significance of ventilation was approved by a super
spreading event in a Hong Kong hospital (Qian & Zheng, 2018). Another recent case is COVID-19, a global
pandemic. According to the research of EPA (the United States Environmental Protection Agency),

13
airborne particles and droplets are the major spreading medium of COVID-19 (EPA, 2022). People who
are infected can release such particles and droplets with SARS CoV-2 virus into the air through exhaling
activity, such as quiet breathing, speaking, singing, exercise, coughing, and sneezing. Because of this
property of air contacting transmission, the particles from an infected person can move in the entire space
especially the room with insufficient ventilation. When it comes to the outside space, a reasonably safe
distance between people has been determined to be six feet to prevent unnecessary interaction with other’s
respiratory droplets (Indoor Air and Coronavirus (COVID-19) | US EPA, n.d.). Although the risk of
infection decreases with time and longer distances between infected people, indoor environments with
inadequate ventilation and activities are still much more dangerous than outdoors. Therefore, ventilation
plays an essential role in improving Indoor Environmental Quality by reducing virus spread. However, the
concentration of the virus is hard to detect and analyzed by number. It is suggested to use PM2.5 to record
the virus’s spread because PM 2.5 is indicated as a potential carrier of COVID-19 (Nor et al., 2021).
Additionally, ventilation is related to building energy performance. Proper ventilation can not only improve
indoor air quality but also achieve energy conservation. Too much ventilations will reduce the effectiveness
of the air conditioning system depending on the seasons, while insufficient ventilation will lead to the lack
of fresh air and overrunning of the mechanical system. In this case, this study will contribute to improving
the effectiveness of the ventilation system, by developing IEQ prediction models based on human
physiological responses.
1.4 Summary
As discussed above, quantifying environmental and physiological benefits is key element in many domains.

14
Therefore, this research will focus on the relationship between indoor environmental quality and human
physiological responses in an office building and will establish a IEQ prediction model for an automatic
ventilation system based on the function of physiological responses. The indicators of indoor environment
quality include temperature, RH, CO2, PM2.5, acoustic, lighting, and thermal satisfaction. The human
physiological responses will be recorded by wearing sensors that can dynamically collect heart rate, skin
temperature, electroencephalography (EEG), and electrical activity of the skin (EDA). These indicators will
contribute to establishing a predicted model to improve indoor comfort and reliability. As a result, this study
will reveal technical methods to estimate human comfort and satisfaction by exploiting modern advanced
sensing technologies and existing control infrastructure.

15
CHAPTER 2: BACKGROUND AND LITERATURE REVIEW
There is much research discussing indoor environmental quality and human physiological responses. This
chapter reviews many existing studies for having a better understanding of the research topic. Based on the
significance of IEQ, there are four main parts: The impact of indoor environmental quality, The assessment
of indoor environmental quality, the correlation between Indoor environmental quality and human
physiological responses, and Data analysis approaches in human responses and the IEQ domain.
2.1 The impact of indoor environmental quality
Indoor environment quality has many impacts on human lives. Identifying the influence of IEQ is the first
step to clarifying the research target. The following research investigated IEQ’s effects in various aspects.
There are many studies focusing on the impact of IEQ in occupant well-being and comfort. Yousef et al
(2016) have a review of literature about it (al horr et al., 2016). According to ASHRAE guidelines, since
people spend 80% to 90% of their time indoors, much research suggests a range of health and comfort
problems are related to Indoor Environmental Quality (ASHRAE, 2016).
Regarding the hospital environment, it is discovered that there are mainly two ways of the influence of IEQ:
mental and physical complaints (Sadek & Nofal, 2013). For physical health, the temperature and air quality
of facilities impact occupants’ health conditions. Additionally, the presence of indoor air pollutants leads
to a high risk of infections or death in patients (Nimlyat et al., 2021). Ramaswamy et al. (2010) investigated
that because of the bad control of air contamination and fresh air, patients are infested with diseases other
than what they have already been treated. Therefore, it is concluded that contaminants and dangerous

16
substances should be adjusted to a certain level by sufficient ventilation which can provide fresh air
(Ramaswamy et al., 2010). When it comes to mental complaints, patients may worry about illness, fear of
medical procedures, and even stress and anxiety because of the sounds and smell of the hospital (Nimlyat
et al., 2021).
Regarding the residential and commercial indoor environment, thermal comfort has a direct link to energy
consumption, since any extent of discomfort leads to modifying the control system to optimal levels.
According to Yang et al. (2013), when the HVAC system applied consistent setpoint temperature control
compared with another system using the adaptive model to dynamically control setpoint temperature, the
latter one shows substantial energy savings for both office and family houses. In detail, a 6% reduction was
achieved in HVAC electrical consumption in an Australian office building and a 33% energy saving of total
energy cost in a hot desert area in Riyadh (L. Yang et al., 2014). Therefore, thermal comfort control has a
large impact on building energy. Additionally, acoustics and lighting comfort is determined as significant
factors of occupants’ productivity (Nimlyat et al., 2021). According to Shengxian et al. (2022), there are
five types of offices affected by acoustics and lighting: cell offices, shared-room offices, open-plan offices,
flex offices, and combo offices. For example, low noise levels and high privacy are more important to
employees in the open-plan office while people in cell offices pay more attention to adequate lighting and
comfortable furnishing (Kang et al., 2022). The experiments took place in an open-plan office in Shenzhen,
China. The results show that work productivity and acoustic satisfaction are both correlated with the
perceived noise level and speech interference because these acoustic problems affect people’s
reconcentration and problem-solving speed. Furthermore, Wolkoff et al. (2021) investigates some indoor

17
air quality indicators’ impact. When it comes to the influence of air temperature, it is suggested that higher
temperature makes the emission rates of chemical gaseous compounds more active compared to lower
ambient temperature. Also, eyes and airways will be more vulnerable because of lower indoor air humidity,
since desiccation causes less efficient mucociliary clearance. Under this condition, the mucous membrane-
related symptoms, such as dry and tired eyes, are more likely to appear and deteriorate working or living
quality (Wolkoff et al., 2021). Thus, people are more inclined to feel the air is not good if the air is humid
or hot (Higuera-Trujillo et al., 2017).

2.2 The assessment of indoor environmental quality indicators
There are many different indicators of Indoor Environmental Quality with various categories of methods.
For example, Ganesh (2021) suggests that acoustic, visual, and thermal comfort are three categories when
investigating indoor environmental quality (Ganesh et al., 2021). From these three parts, environmental and
personal factors are summarized, and researchers utilize these factors as indicators to show the change in
an indoor environment. Environmental factors include air, radiation and operative temperature, air
movement, relative humidity, contamination, and climate conditions. The typical personal factors are
metabolic rate, clothing insulation, and thermal history (Ganesh et al., 2021). All these indicators are
considered as the research parameters of IEQ to perform stable and reliable results. In order to understand
the IEQ components and measurement, the following part reviews many assessments of indicators in four
components of IEQ: Indoor Air Quality, Thermal Comfort, Lighting, and Acoustic.

18
2.2.1 Indicators of Indoor Air Quality current technologies
There are a large number of IAQ indicators including CO2, PM, and hundreds of chemical components
present in the indoor air. Cony Renaud Salis et al (2017) selected some pollutants which are known as
harmful and important for human health to measure indoor Air Quality, including acetaldehyde, acrolein,
α-pinene, benzene, carbon dioxide, formaldehyde, naphthalene, nitrogen dioxide, PM10, PM2.5, radon,
styrene, toluene, trichloroethylene, TVOC, and mould (Cony Renaud Salis et al., 2017). According to
Sadrizadeh et al (2022), for adults and children, indoor contaminants, such as CO 2, PM, VOCs, NO x, and
ozone, are recognized as the factor causing severe health problems. Therefore, it is significant to identify
IAQ indicators for providing healthy Indoor Air Quality.
In detail, VOC pollutants are the common emission from construction materials, furnishing, resins of wood
products, adhesives, cleaning chemicals, and carpets in the indoor environment (Sadrizadeh et al., 2022).
The concentration of VOCs includes the measurement of formaldehyde, benzene, toluene, naphthalene, and
xylene. For evaluating the impact of VOCs, a questionnaire study was developed for eight schools in
Sweden and the results revealed VOCs are related to the presence of asthmatic symptoms (J. L. Kim et al.,
2007). When it comes to CO 2, Madureira et al. (2015) indicated that the concentration of CO 2 is a significant
parameter for IAQ evaluation. The experiment took place in a primary school in Portugal. When CO 2
concentration exceeded 1000 ppm, Indoor Air Quality decreased, while IAQ is better in the situation CO 2
concentration is lower than 1000 ppm (Madureira et al., 2015). In terms of PM, PM exposure is recognized
as an IAQ indicator. PM 10 particles enhance the risk of respiratory sickness and lung cancer and when PM 2.5
is in the range of 20.5 ±2.2 mg/m3, there is more chance of conjunctivitis, hay fever, an itchy rash, and

19
sensitization to outdoor allergens. PM 2.5 also shows relationships with virus infections. According to Ayoub
Meo et al. (2021), the measurement of PM 2.5 took place in London and the COVID-19 data about new cases
and death was based on UK-govWeb. The results presented a positive correlation between PM2.5
concentration and COVID-19 cases (Ayoub Meo et al., 2021). Similarly, Renard et al. (2022) did an
analysis of the correlation between PM2.5 and COVID-19 cases in 32 locations in 6 countries, including
France, Germany, Italy, Netherlands, Spain, and the United Kingdom for the 2020-2022 period. After data
analysis, it is suggested that the concentration of PM 2.5 has a strong indirect effect on Covid-19 mortality
(Renard et al., 2022). Therefore, PM concentration plays a significant role in indicating the virus in IAQ.
Furthermore, Sun et al. (2021) investigated the correlation between indoor air pollutants and children’s
health by testing several indicators. CO 2 is the first indicator tested by HUMLOG 20 made in Austria.
Formaldehyde is tested by Formaldehyde Multimode Monitor from Japan. RING passive adsorption
sampling tube is used to detect VOCs concentration. PM 2.5 and PM 10 are tested by Aerosol Mass Monitor
831. Microflow-α from Italy is utilized for measuring Fungi. The dust sampling method is used for dust
detection by MC-DL202(Sun et al., 2022).

Number Indicators Equipment
1 CO 2 HUMLOG 20, E + E Inc, Austria

(Data Logger for Humidity, Temperature, Air Pressure and CO2 in
HVAC Applications., n.d.)
2 Formaldehyde Formaldehyde Multimode Monitor, SHINYEI Inc, Japan

20

(S E N S O R L I N E U P, n.d.)
3 VOCs RING passive adsorption sampling tube

4 PM 2.5 and PM 10 Aerosol Mass Monitor 831, Met One Instruments, Inc, USA

(Aerocet 831 Handheld Aerosol Mass Monitor - Met One Instruments,
n.d.)
5 Fungi Microflow-α, AQUARIA Inc, Italy

(Active Sampler for Microbiological Air Sampling “Microflow Alfa” |
Aquaria Srl, n.d.)
6 Dust sampling MC-DL202, Panasonic Inc, Japan
Figure 2.1 IAQ indicators and measurement (Sun et al., 2022)
2.2.2 Indicators of Thermal Comfort existing model
Thermal comfort is an expression of human satisfaction that is affected by many environmental indicators,
such as ambient temperature, mean radiant temperature, air velocity, humidity, solar radiation, clothing
insulation, metabolic rateand skin temperature. Tang et al. (2021) investigated the relationship between

21
health risk and thermal parameters by analyzing the thermal satisfaction of workers. The results suggested
that ambient temperature and wind speed greatly impact workers’ satisfaction (Tang et al., 2021). Song et
al. (2021) researched the effect of solar radiation on indoor thermal comfort. It is suggested that the impact
of solar radiation on thermal comfort is more server in cold weather (B. Song et al., 2022). In order to
integrate the effects of these indicators on thermal comfort, some models are developed for having a
comprehensive evaluation of thermal comfort. Predicted mean vote (PMV), Draft rate evaluation, and
Predicted percentage of dissatisfaction (PPD) are the most common and popular models for thermal comfort.

Figure 2.2 The main factors impacting thermal comfort (Guenther, 2021)
Firstly, Fanger’s Predicted Mean Vote PMV model is the most classical thermal comfort model so that it is
widely applied in most international and national standards regarding air-conditioned buildings, such as
ASHRAE 55-2020, ISO 7730, and GB 50736-2012 (Du et al., 2022a). The PMV index is on a seven-point
scale, which is used to predict the mean value of votes of a group of occupants in the indoor environment
based on the heat balance of an individual. It considered two kinds of parameters combining together in the
PMV index: one is environmental parameters including air temperature, air velocity, mean radiant

22
temperature and relative humidity; another one is occupants’ parameters including metabolic rate and
clothing insulation (Benabed & Boulbair, 2022a).

Figure 2.3 A seven-point scale of PMV (Guenther, 2021)
Secondly, Draft rate (DR) evaluation is the model focusing on air movement and air velocities. Because it
is common that people at low activity levels are sensitive to air velocities, Draft is defined by ASHRAE to
identify unwanted local cooling of the body caused by air movement. In detail, Draft rate can be evaluated
by the effects of airflow velocity, turbulence intensity, and air temperature (Benabed & Boulbair, 2022b).
Compared with PMV, DR is inclined to represent the local thermal dissatisfaction resulting from the Draft
(Tian et al., 2022).
Thirdly, the Predicted Percentage of Dissatisfied (PPD) can be derived from PMV to predicts the percentage
of people who prefer feeling too warm or too hot (Du et al., 2022b). PPD’ development is from climate
chamber experiments and the utilization of this model is a priority in an air-conditioned environment (Du
et al., 2022b). Additionally, the contributors of PPD include drafts, abnormally high vertical temperature
differences between the ankles and head, and/or floor temperature (Guenther, 2021).
2.2.3 Indicators of Lighting and Acoustics
Lighting and Acoustics are two significant components of IEQ evaluation. There are many studies focusing
on measuring the indicators of lighting and acoustics to investigate the correlation with indoor comfort.
For lighting, the basic qualities are intensity, form, color, direction, and movement. Illuminance is the

23
metric for measuring lighting intensity within an area. The units include footcandles (IP unit) or lux (SI
unit), which are determined by the amount of light (lumens) per square foot or meter (How to Measure
Light Intensity: Understanding & Using a Lux Meter, n.d.). Deng et al. (2021) investigated indoor lighting
quality for predicting work engagement by measuring three different lighting levels’ environments and the
results found that productivity is related to indoor lighting conditions but has discrepancies between
different people. According to Lee et al. (2014), it is concluded that indoor lighting levels impact visual
responses and the mood of occupants by testing occupants for 500 lx and 750 lx illuminance levels under
3,000 K, 4,000 K, and 6,500 K conditions (Lee et al., 2014). Additionally, Hwang and Kim (2011) used
questionnaires and self-reports to evaluate lighting satisfaction. The results conduct that lighting problems
including glare, darkness, unqualified materials, and logical error of shade lead to visual annoyance (Hwang
& Jeong, 2010).
Regarding acoustics, the noise level in decibels (dB) is the common indicator in acoustics studies. It is
important to achieve acoustics assessment for indoor environment prediction. Lam et al. (2022), who did
research in a Singapore hospital, pointed out that all occupants’ groups including patients and staff
expressed noise level as the most direct factor of dissatisfaction among sixteen examined IEQ factors.
Notably, according to Cain et al. (2013), although high noise levels can reduce people’s concentration and
effectiveness, loudness can be even desired and required in some indoor cases (Cain et al., 2013).

2.3 The correlation between Indoor environmental quality and human physiological responses
According to the Adaptive Comfort theory, occupants are active responders interacting with the indoor

24
environment rather than passive objects (Wu et al., 2019). That means although people have different
physiological histories and different acceptable environmental ranges, occupants can adapt to the indoor
environments through physiological, psychological, and behavioral adaption(Wu et al., 2019). Therefore,
there are many previous studies that investigated the correlation between IEQ and human physiological
responses and resulting in many complex and thought-provoking conclusions.
IAQ, as the first component of IEQ, is investigated with physiological responses by many studies. Kim et
al. (2017) developed an integrated physiological response score for occupants based on 22 experience
participants’ activities under different IEQ conditions. The experiments took place in three scenarios during
an 8-hour working period. IAQ satisfaction survey and thermal comfort questionnaires were conducted
during these three scenarios. The results show the occupants were more impacted by the operative
temperature than CO 2 concentration, and indoor air pollution is one of the factors affecting mental health
(J. Kim et al., 2017). Additionally, Kim et al. (2018) conducted more detailed relationships between IAQ
factors and human responses in the same experiment. When CO 2 concentration is lower than 1000 ppm,
physiological responses maintained stable conditions under work stress. When CO 2 concentration is higher
than 2000 ppm, the subjects are difficult to maintain homeostasis under the same work stress(J. Kim et al.,
2018). Therefore, the CO 2 level is one of the most impactful indicators of IAQ. In terms of temperature,
Yang et al. (2022) conducted experiments on physiological responses in the urban and rural residential
environment under three temperatures, 18 ℃, 14℃, and 8 ℃. The results show skin flux and skin temperature
are two physiological parameters showing significant changes in these three environments(R. Yang et al.,
2022).

25
Regarding thermal comfort, Wu et al. (2022) assessed thermal comfort and physiological responses among
twenty participants under three environmental conditions. Blood oxygen saturation (SpO2), skin
temperature, and electrocardiograph (ECG) were measured for physiological responses, and questionnaires
are applied for thermal comfort evaluation. It is concluded that thermal sensation becomes quicker than
physiological responses to be a stabilization condition (Wu et al., 2022). Another experiment made by
Mansi et al. (2022) used two non-invasive wearable devices to measure electroencephalography (EEG),
Heart Rate Variability (HRV), electrodermal activity (EDA), and skin temperature (ST) on 52 participants
under different thermal conditions (cold, warm, and neutral). The results showed that physiological signals
can identify cold and warm environments uniquely, while neutral sensations are less distinguishable (Mansi
et al., 2022).
When it comes to lighting and acoustics, Juan et al. (2022) utilized a portable brainwave instrument to
capture brainwave signals in different indoor environments. Lighting and sound are two of the significantly
correlated parameters with the participants’ concentration levels and anxiety levels (Juan & Chen, 2022).
Additionally, lighting performed various impacts on individuals’ work engagement measured by
electroencephalography (EEG) in the experiment under three typical lighting levels, 200 lux, 500 lux, and
1000 lux (Deng et al., 2021).

2.4 Current technology strategies in smart ventilation control for energy efficiency
Considering the impact of ventilation on energy saving, there are many advanced technology strategies in
the smart ventilation control system. Smart ventilation is a continuously adjustable process to provide

26
desired IAQ and thermal benefits as well as energy savings to be responsive to the responses of occupants,
outdoor calvin Liu environments and operation of air movement and air cleaning requirements (Durier et
al., 2018). Figure 2.4 shows the main features and responsible parameters of smart ventilation.

Figure 2.4 Main features of smart ventilation (left) and responsible parameters for smart ventilation
system(Durier et al., 2018)

2.4.1 Benefits and limitation of smart ventilation system
There are mainly 4 benefits of the smart ventilation system. First, the throttle control is occupancy-based.
Different kinds of buildings need different ventilation strategies. For example, a building filled to the brim
requires maximum ventilation, while ventilation is not needed for a vacant building. For the building with
fluctuated occupancy, multiple ventilation settings should be applied throughout one day. In terms of this
situation, operators had to manually adjust the ventilation system by self-detecting in the past. Nowadays,
smart ventilation systems can perform automatically by using occupancy-cased throttle control based on

27
the sensor’s judgment about the change of the number of occupants(Durier et al., 2018). Second, the smart
ventilation system can optimize the building’s indoor environment to adapt to the next day’s weather based
on the weather forecast(Durier et al., 2018). If the weather for tomorrow is expected to be one of the hottest
days of the summer, the smart ventilation system will prepare to start to cool the building from the night.
Besides hot days, it contains various adaptation mechanisms for rainy, stormy, snowy, etc. Third, energy
saving in smart ventilation systems is a highlighting benefit(Durier et al., 2018). According to
manufacturers of smart ventilation control systems, 20% to 30% of building operational energy can be
saved by using smart vents (Best Smart Vent: Room by Room Climate Control, n.d.). Furthermore, it can
reduce CO 2 concentration because of saving energy. Almost 39% of global CO2 emissions are contributed
by old buildings (Durier et al., 2018). The main reason is their overconsumption of energy by using dated
HVAC. This dated HVAC has no access to automatically control the environment according to occupants
or weather conditions. At last but not least, the smart ventilation system achieves remote control; and real-
time monitoring. The remote-control function solves the problems when customers forgot to turn off the
HVAC before going outside. The real-time monitoring technology adds visibility and predictability to
indoor usage statistics so that future energy consumption and customer living habit is available to forecast
helping in budgeting and identifying avenues of saving (Durier et al., 2018).
However, people’s physiological responses to environments are different because of different thermal
histories. The smart ventilation system treats occupants as the same subject and does some adjustments so
that it will lead to some dissatisfaction from part of occupants. Therefore, it is helpful to develop the
correlation between human physiological responses and indoor environments to reform a better smart

28
ventilation system.
2.4.2 Applications of smart ventilation
There are many current products utilizing smart ventilation to achieve energy savings. Dedicated Outside
Air System (DOAS) is one of the good examples applied smart ventilation. DOAS is an HVAC unit that
can bring outside air into the indoor environment independently from the heating and cooling system
(Nguyen, 2016). Compared with the central VAV system with zones, DOAS delivers the required
ventilation air more accurately. Additionally, it can configure to control dewpoint and temperature setpoint
providing good humidity and thermal condition(Dedicated Outside Air System DOAS - MEP Academy,
2022).
The personal environmental control system is also a popular tool to achieve energy efficiency and occupants’
customization. In order to have a better understanding about current techonologies of the ventilation
strategy focusing on human physiological responses, personal ventilation is an example applied the human
feelings into the product. Personal ventilation as a kind of smart ventilation focusing on human bio-signals
has already been practiced from research to usage (Melikov et al., n.d.). Figure 2.5 shows a developed
prototype with air terminal devices (ATD) for open working station. The ATD is attached to the arm of the
occupant rotating around its vertical axis. The direction of the supplied personalized flow can be changed
in the vertical plan. In this way, the occupants can customized the personalized air supply to face, chest or
any other angle. However, this technology is still not very convenient for occupants to wear because of the
connection with the human body.

29

Figure 2.5 Personalized ventilation in use (Melikov et al., n.d.)

2.5 Data analyses approaches in human physiological responses and IEQ domain
Data analysis is a popular approach when establishing relationships between various parameters. It is
applied in the human physiological responses domain to resolve the data of questionnaires and
measurements. There are many studies using different statistical software and methods reviewed in this part.
Wu et al. (2022) assessed thermal comfort, temperature step changes, and physiological responses to
investigate the relationship between human comfort and physiological mechanism (Wu et al., 2022). SPSS
and R studio software were used to process data. There are three test methods for different datasets: the
Shapiro-Wilk test is for analyzing the normality of data; the repeated-measures ANOVA or paired-sample
t-test determined distributed datasets; abnormally distributed datasets were processed by the nonparametric
test-Wilcoxon's test or Friedman test (Wu et al., 2022). Besides using multiple test models in SPSS and R
studio, machine learning is applied for developing correlations between IEQ and human physiological
responses. Torku et al. (2022) detected the interaction of environmental indicators and older adults’

30
physiological sensations by using machine learning (Torku et al., 2022). In detail, older adults’
physiological reactions were monitored by wearable sensors, and the global positioning system (GPS) was
used to record environmental data during the outdoor walk. Three approaches were applied to test and train
the correlation between their physiological signals and environmental data: Machine learning algorithms
including Gaussian Support Vector Machine, Ensemble bagged tree and Deep belief network (Torku et al.,
2022). Gaussian Support Vector Machine (SVM) refers to a representation of different classes in a
hyperplane, which is used for classification and regression problems (Bandgar, 2021). Ensemble bagged
tree is a kind of machine learning algorithm that uses combined models with decision trees to reduce
variance and prevent overfitting (Kurama, 2019). Deep belief network is another algorithm that can
resemble a deep neural network with a deep architecture (Kalita, 2022). Results indicated that Ensemble
bagged tree algorithm has the highest accuracy which is 98.13% for detecting stress and non-stress samples
and 98.25% for detecting low and high-stress samples. The detected risk stress hotspot locations correspond
to the actual environmental conditions (Torku et al., 2022). Additionally, Kumar Ojha et al (2018)
investigate urban environments on human physiological responses by using machine learning. Wearing
technology and environmental sensors are also the main device for collecting human responses and urban
space conditions (Ojha et al., 2019). Machine learning techniques with classification, fuzzy rule-based
inference, feature selection, and clustering were used to develop the relationship between indicators and
predictive models (Ojha et al., 2019).

31
2.6 Summary
In this chapter, there are many literatures reviewed about the impact of indoor environmental quality, the
assessment of indoor environmental quality indicators, the correlation between Indoor environmental
quality and human physiological responses, current technology strategies in smart ventilation control for
energy efficiency, and data analyses approaches in human physiological responses and IEQ domain. These
papers fulfilled the background information and identified the significance of the research target.
Additionally, although there are many papers investigate human bio-signals and Indoor Environmental
Qualities separately, few studies focused on using human bio-signals to predict Indoor Environmental
Qualities based on their correlations. Therefore, this research was intended to fill this gap by using advanced
correlation analysis and machine learning.

32
CHAPTER 3: METHODOLOGY
Some basic research backgrounds and investigation precedents have been developed in the first two
chapters. In the Methodology, there are 4 parts including the project scope, experimental procedure, data
collection, and data analysis approach.
3.1 Project scope
The goal of the research is to improve the ventilation control system by using human physiological
responses (bio-signals) as indicators to predict Indoor Environmental Quality (IEQ). In order to develop
the predicted model for the ventilation control system, it is desired to exploit correlations between Indoor
Environmental Quality (IEQ) and human physiological responses (bio-signals). Therefore, the experiments
focus on exploring the more relevant bio-signals to the IEQ parameters based on the collected data to
develop prediction models by using Machine Learning. Figure 3.1 describes the progress of the research.

Figure 3.1 Whole progress
According to Minghuan’s work, the research has already developed in the residential area. In order to follow
the previous work, data are collected with same sensors in two commercial spaces. The same types of data
are focused, including indoor and outdoor environmental Quality parameters and participants’
Collecting
data in
experiments
Investigating
the
correlations
Determining the more
relevant bio-signals
Developing the
Prediction
models
Actuator
and
operations

33
physiological responses. Environmental Quality parameters consist indoor air temperature, indoor humidity,
outdoor PM 2.5 concentration, CO 2 concentration, acoustics, and lighting. Besides, two surveys are provided
for thermal condition and air quality satisfaction. In terms of the occupants’ bio-signals, it comprises heart
rate, skin temperature, stress level and EDA.

3.2 Experimental procedures
The experiments take place in the LA Downtown office, and the USC School of Architecture Watt Hall working
studio. The DT office area is divided into 2 zones according to the mechanical zoning plan. The experiments are
implemented in the right zone which faces southeast. Figures 3.2 and 3.3 show the floor plan of DT office and
USC School of Architecture Watt Hall working studio. The green area is where the experiment is done.

Figure 3.2 DT office area

34

Figure 3.3 USC School of Architecture Watt Hall Working Studio
In DT office, there were 4 volunteers (3 females and 1 male) involved in the experiment. In the USC School
of Architecture Watt Hall working studio, 5 males and 5 females participated in this study. The range of
age is from 22 to 35. All the participants were healthy and had no history of asthma. In each experimental
spot, volunteers were required to work or study in the office as usual for 1-3 days and 4 to 6 hours were
spent for each experimental day. Before the experiments, the sensors are installed in the middle of each
experimental space for collecting indoor and outdoor environment quality parameters during the experiment
and uploading data to the cloud data platform. Additionally, because these two experimental sites had no
operational windows which means they were closed environments, there was a high probability that the

35
concentration of indoor PM 2.5 is approximately 0. Considering the building infiltration, the outdoor PM 2.5
could somewhat reflect the indoor PM2.5 situation. Therefore, outdoor PM 2.5 sensors were installed on the
outdoor balcony to guarantee enough environmental PM 2.5 data could be used. Besides, each participant
wore two smartwatches for collecting bio-signals of human physiological responses, including heart rate,
skin temperature, EDA, and stress level. In the meantime, two surveys about the assessment of indoor
thermal comfort and indoor air quality satisfaction were filled in every two hours during work or study time.
The time interval for collecting IEQ and bio-signals were all 10 minutes except for the illuminance sensor.
For the lighting conditions, because these two spaces were usually turned on artificial light during the whole
day, the illuminances didn’t change too much. Therefore, lighting sensor tested in three specific time for
different impacts of natural lighting: 10 am, 1pm and 5pm. Table 3.1 shows a summary of sensors and
smartwatches used in the experiment.

Devices in the
experiment
Measured indicators’
category
Indicators
HOBO MX 1102

Indoor Environmental Quality
Air temperature, Humidity, and CO 2
concentration
Pa- II- SD Particulate Matters concentration
(PM 2.5 and PM 10)
Dr.Meter LX1330B
Digital Illuminance Light
Meter
Illuminance level
PCE-SDL 1 Acoustics level
Garmin Vivosmart 3 Human Physiological
responses
Stress level and Heart rate
Empatica Embrace 2 Skin temperature and EDA
Table 3.1 Experiment devices

36
After experiments for data collecting, the statistic software (Minitab) and machine learning (Python) were
applied for data analysis to identify correlations between these parameters, including the correlation
between IEQ parameters and human bio-signals, the relationship between IEQ satisfaction and human bio-
signals, and the relationship between thermal comfort and bio-signals. Furthermore, common features of
physiological responses across all the participants in their indoor environmental quality conditions were
found, and the more relevant bio-signals were determined. Figure 3.4 shows the Methodology diagram.
There are three stages: the preparation stage, the experimental stage, and the data analysis stage. In the
preparation stage, IEQ and human physiological response indicators were identified for future experiments’
measurements through the literature review. During the experiments, there were totally 14 volunteers did
work and studied in the DT office and USC MBS studio. Meanwhile, researchers measured identified
indicators of IEQ and bio-signals by utilizing sensors and smartwatches as well as IEQ and IAQ satisfaction
through surveys. Finally, Minitab and Python were used in the data analysis stages for data organization
and correlation analysis. After identifying the correlations, predicted models for smart ventilation strategy
were generated to predict IEQ using bio-signals as indicators.

37

Figure 3.4 Methodology Diagram

38
3.3 Data collected
In order to collect enough data for analyzing the relationships, there are three types of information needed:
indoor environmental quality conditions, participants’ bio-signals, and survey results about participants’
thermal comfort and IAQ comfort.
3.3.1 Data for Indoor Environmental Quality
IEQ parameters are used for identifying the current state of indoor or outdoor environment quality, such as
lighting, acoustic, and air quality. According to ASHRAE, workstations illuminance level (lux), acoustic
decibel (dBA), indoor temperature (℉/℃), PM 2.5 (μg/m^3), and CO 2 level (ppm) are recommended to
record for building environments. During the experiments, multiple sensors were used to collect Indoor and
Outdoor Air Quality, acoustics, and lighting conditions. Regarding Indoor Air Quality, air temperature,
humidity, and CO 2 concentration were the main indicators to estimate the correlations by HOBO MX 1102.
PM2.5 concentration was tested by Pa-II-SD. Acoustics levels were measured by PCE-SDL 1. Dr.Meter
LX1330B Digital Illuminance Light Meter was used to record lighting conditions.

 HOBO MX 1102 - Air temperature, Humidity, and CO 2 concentration
To measure indoor temperature, relative humidity, and CO2 concentrations, the HOBO MX 1102 was
chosen. This sensor has a USB connector and phone app with Bluetooth, which allows it to be connected
to a computer and phone running HOBO ware to set the settings for measuring data. Once the parameters
have been set, the data can be read and downloaded in excel format. The period between data collection
was ten minutes, and all collected data were automatically saved in memory. The battery of the HOBO

39
sensor can last for at least a year and the data storage capacity is about 176 days continuously.
HOBO sensors were installed every day before starting the experiment and data of the day were downloaded
through the HOBO app after finishing the experiment just in case there is any unexpected data storage
emergency. When the HOBO sensor runs the measurement, the screen shows the instant data. When the
HOBO sensor stops running, STOP sign appears on the screen. Researchers paid attention to the HOBO
screen to guarantee the operation of HOBO sensors during the experiment.

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
Table 3.2 HOBO MX1102 Details

40

Figure 3.4 HOBO in experiment
 Pa- II- SD – Particulate Matters concentration (PM 2.5 and PM 10)
The Purple Air Company's Pa- II- SD air quality sensor was used to collect the PM 2.5 data; it counts
particles in certain size ranges and then utilizes a calibration algorithm to estimate PM2.5 mass
concentration for residential, commercial, or industrial applications in real-time. Additionally, this air
quality detector's built-in Wi-Fi enables it to send data to a PurpleAir map and save that data for usage by
any smart device. Additionally, this sensor can incorporate an SD card and a real-time clock to record and
save data locally in places with limited or no Wi-Fi access. The study used a ten-minute measurement
window for its data.
Before the experiment, PA-II-SD sensors were installed both indoors and outdoors to record the PM 2.5 and
PM 10 concentrations by plugging in the inside and outside sockets. Researchers logged in the sensors’
location and product serials number to make sensors show on the PurpleAir map website to monitor the
simultaneous data. After the experiment, the data were downloaded from the PurpleAir website through
API method.

41
Technical 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
Table 3.3 Pa- II- SD Details

Figure 3.5 Pa-II-SD in experiment

 Dr.Meter LX1330B Digital Illuminance Light Meter – Illuminance level
For measuring the indoor illuminance data, a Dr.Meter LX1330B Digital Illuminance Light Meter was used.
However, it lacks a storage feature. In this study, the researchers will measure the illuminance data multiple
times in a single day (morning, afternoon, night).

42

Technical 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
Table 3.4 Dr.Meter LX1330B Digital Illuminance Light Meter Details

Figure 3.6 Dr.Meter LX1330B Digital Illuminance Light Meter in experiment
 PCE-SDL 1 – Acoustics level
The PCE-SDL 1 sound level data logger, which is intended for noise, quality control, and all types of
ambient sound measurement, was used to measure sound level for the purpose of gathering acoustic data.
Acoustic measurements were taken every ten minutes. Sound datalogger software should be downloaded
to read the data provided by sensors. In the experiment, the sensors were set in the laptop for 10 minutes
measurement interval before starting. After the whole day’s test, the data were downloaded by Sound

43
datalogger software in excel file.

Technical 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
Table 3.5 PCE-SDL 1 Details

Figure 3.7 PCE-SDL 1 in the experiment

3.3.2 Data for Surveys
The perception of indoor environmental quality among the occupants will be investigated using two surveys.
One of the surveys is how the residents feel about the quality of the indoor air. Another survey asking about

44
human thermal comfort is implemented in this experiment for assessing temperature and humidity, which
is related to human thermal comfort. Each participant will receive these two surveys in paper. Additionally,
individuals had to fill out forms requesting their age and gender before to their experiment. To increase the
accuracy of the evaluation, a seven-point scale will be used to gauge how the occupant feels about the
quality of the indoor air, and how they feel about their thermal comfort. Figures 3.8 and 3.9 show the
surveys’ details.
However, it was difficult to gather user accurate data due to the fact that participants in this study must
complete questionnaires every two hours for three days in a row. Survey data collection largely relied on
user awareness and researcher reminders because participants occasionally forgot to complete the
questionnaire at the required time point. Therefore, researchers set alert clocks every two hours to remind
them. Participants were given two paper surveys regarding their opinion of the air quality and their level of
thermal comfort (-3, -2, 1, 2, 3, 0, 1, and 3).

45

Figure 3.8 Indoor Air Quality Evaluation survey

46

Figure 3.9 Indoor Thermal Comfort Evaluation Survey

47
3.3.3 Data for participants’ physiological responses
There are two smartwatches for collecting human bio-signals: Garmin Vivosmarts 3 and Empatica Embrace
2. Garmin Vivosmart 3 can record stress level and heat rate. Empatica Embrace 2 is used for collecting skin
temperature and EDA.

 Garmin Vivosmart 3 – Stress level and Heart rate
The experimenter's heart rate and level of stress were recorded in real-time using a Garmin Vivosmart 3
device. The battery for Garmin Vivosmart 3 can last at least one week so that it is not needed to charge
every day. The data were stored in the Garmin Connect personal account for each user. Therefore,
participants are required to sign into a new account in Garmin Connect for getting data access. Before
starting, the Garmin Connect app was requested to download on the smart phone for logging in the new
account. During the experiment, volunteers needed to synchronize real-time data on the phone to the
Garmin connect user website. However, the default time interval between the two data collections was not
10 minutes and the files were FIT instead of excel file. Researchers did further data processing for stress
level and heart rate to summarize the time interval to 10 minutes and transfer the data files into excel.

Technical details
Measured parameter Heart rate; Stress level
Range 0~200 bpm; 0~100
Resolution 1bpm; 1 unit of stress level
Table 3.6 Garmin Connect Details

48
 Empatica Embrace 2 – Skin temperature and EDA
Empatica Embrace 2 was used to collect real-time data of the experimenter's electrodermal activity (EDA)
and skin tempertaure. Embrace Reasearch portal was used to access users’ data. Meanwhile, participants
needed to download app Mate for Embrace watch to connect the smartwatch and phone. The data were
stored in the website automatically when it is connected to the app on the phone. Additionally, Embrace 2's
EDA and skin temperature data could be accessed and downloaded for a $600 monthly subscription on
Embrace Research's website. The time interval of two data collection was also not 10 minutes. A Python
script was also required to convert the enormous amount of data downloaded from Embrace Research into
data that could be read because the time was measured in milliseconds. Comparing with Garmin Vivosmart
3, Embrace 2’s battery is not as powerful as it so that Embrace 2 need to be charged every day after the
experiment.

Technical 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℃; /
Table 3.7 Empatica embrace 2 Details
Figure 3.10 shows participants wearing two smartwatches together on the left arm during the experiment.

49

Figure 3.10 Smartwatches in the experiment
3.4 Data analysis approach: Cross correlation analysis and Machine Learning
Following data collection, it was necessary to consolidate all the information for each participant in a single
Excel file, with each entry reflecting the same time. Data of some indicators should be recalculated by
finding the average or median to get more accurate results standing for a 10-minutes period. And the data
of human physiological signals were missed during the data collection process sometimes since the
participants did not stay in the working area for the entire time. Therefore, the missing data needs to be
cleaned before data analysis.
After categorizing the data, Python was the main tool used into the further analysis. The Time Lagged
Cross-Correlation (TLCC) & Windowed TLCC method were used in the first step to illustrate the
correlations between Bio-signals and IEQ indicators. Granger causality test was applied to verify the
significance levels of the developed correlations. In order to filter correlations for establishing prediction
model, TLCC and Granger causality test worked together to select the correlations which had reasonable
leading relationships and significantly related level by observation of offsets and p values. Some

50
comparisons by genders, and experimental spaces are generated to discuss the findings based on the
developed correlations. Then, machine learning approach was used for establishing prediction models for
individuals’ data and general data.
3.5 Summary
In chapter 3, the research target and aims were demonstrated at first. In order to improve the ventilation
control system by using human physiological responses (bio-signals) as indicators, the predicted model for
the ventilation control system is desired to developed based on exploiting correlations between Indoor
Environmental Quality (IEQ) and human physiological responses (bio-signals). Then, the methodology for
the experiment and data analysis were described with devices and applied technologies. Finally, there are
14 participants involving in the experiment and provide over 16 hours’ data each person for data analysis.

51
CHAPTER 4: DATA ANALYSIS
In Chapter 4, the data organization process is illustrated first to establish the database which is the basement
of all the further steps. Then data analysis methods are introduced including Time Lagged Cross Correlation
(TLCC) & Windowed TLCC analysis to investigate the correlations between human physiological response
indicators and IEQ indicators. After analyzing the correlations, some indicators’ significant correlations
were summarized, but it is not persuasive enough to conclude they can develop the predicted model.
Therefore, granger causality tests were applied for determining whether one bio-signal is useful for
forecasting one IEQ indicator and selecting the final causal-related indicators. Eventually, predicted models
for selected indicators would be established by using Machine learning in Python.
4.1 Data organization
According to the descriptions of utilized sensors and smartwatches in chapter 3, there are 3 different kinds
of problems needed to be solved for further data analysis. Firstly, the time intervals of each indicator should
keep the same, which is 10 minutes. However, only acoustics sensors, PM2.5 sensors, and HOBO sensors
for air temperature, CO2, and RH did the 10-minute data collection. Therefore, data for other indicators
needed to be reorganized. Two smartwatches couldn’t collect data every 10 minutes because of the default
technical settings. For the Garmin Connect smartwatch, it stores stress-level and heart rate data every 1
minute. For the Empatica embrace 2 smartwatch, EDA and skin temperature were collected every 1/4
second.
 Garmin Connect smartwatch - Stress-level and Heart rate

52
The data files downloaded from Garmin connect website were fit Files instead of csv or xlsx files. So
the first step is using Python to change the fit files to xlsx files. Figure 4.1 shows an example of the
stress level and heart rate processed data.

Figure 4.1 Example of stress level and heart rate data
From figure 4.1, the negative stress level values are highlighted with red color because they are
apparently wrong data. The green parts are the available data showing the valid results. After gathering
all the raw data without wrong parts, the data belonging to the time points every ten minutes starting
at nine o'clock are selected for the future data organization, such as the data at 10:40:00.
 Empatica embrace 2 - EDA and Skin temperature
Empatica embrace 2 collected EDA and skin temperature every 1/4 seconds in the csv files. Figure 4.2
shows the excel results for EDA and skin temperature separately.

Figure 4.2 Example of EDA and Skin temperature

53
For the EDA data, the same approach was applied for simplifying results into 10-minute data intervals.
However, the values of skin temperature show an increasing trend. The reason why this happened was
when the smartwatches test skin temperature, skin touch with smartwatches was required. The tested
temperature increased with contacting time passing by. Therefore, in order to make the data more
accurate and avoid this discrepancy because of skin touch, all the data in every 10 minutes were
calculated the average to stand for a certain time period.
Secondly, after downloading the data from sensors, the timestamps for most of the data are Unix timestamps,
which are measured time by the number of seconds that have elapsed since 00:00:00 UTC on 1 January
1970. In order to understand and show the time intuitively, the Unix timestamps should be transferred to
regular time. Thirdly, there are some gaps and errors in the downloaded data. Most of the apparent errors
should be removed and being cautious about the data gaps to avoid data serial problems.
Finally, all the values of indicator for each participant were organized correspondingly into one excel file
which is time-series datasets. Figure 4.3 shows one example of personal dataset.

54

Figure 4.3 Example of organized dataset
4.2 Cross Correlation analysis
After organizing all indicators correspondingly with time, correlation analysis was the next step to primarily
investigate how these IEQ indicators impact and interact with human bio-signals on participants. The used
approach for correlation analysis should take into account adhering to time series data because these data
were a collection of observations gathered through repeated measurements over time (HAYES, 2022).
4.2.1 Applied analysis approaches
 Time Lagged Cross Correlation (TLCC) analysis
Time-lag cross correlation (TLCC) can show the directionality of two signals, such as in a leader-follower
interaction where the leader starts a reaction and the follower repeats it (Jin Cheong, 2019). The correlation
between two signals is regularly calculated while one-time series vector is incrementally shifted to

55
determine the TLCC. The two time series data are at their most synced when the peak correlation is at the
center (offset=0). If one signal precedes another, the peak correlation might be located at a different offset.
Compared with a single correlation coefficient, time lagged cross correlation can avoid the loophole
presented in a possible lead-lag relationship. Prior to calculating the correlation coefficients, this method
constructs both lags and leads of the second variable across the time period while holding one of the series
in place, often the dependent variable. This will allow the comparison of different series against one another
and acquire a more comprehensive knowledge of the pattern. In this way, the lag and lead relationships can
be discussed, which are absent from the single coefficient method, as was previously mentioned (Salami,
2022). Figure 4.4 shows the coding process used for TLCC analysis.

Figure 4.4 Cross correlation function for TLCC analysis
 Windowed TLCC analysis
For Windowed time lagged cross correlations, it is calculated to evaluate the more finely grained dynamics
than TLCC. The time-lag cross correlation is repeated during this procedure in several signal windows.
Following that, we can examine each window individually or add up all the windows to get a score that
compares the differences in how two people interact with one another in a leader-follower fashion (Jin
Cheong, 2019). Figure 4.5 shows the WTLCC analysis in Python. The same cross correlation function was

56
applied in WTLCC as the basic model. Various samples (windows) were set up for the more finely grained
dynamics analysis.

Figure 4.5 WTLCC coding
 Granger Causality Test
The Granger causality test is a statistical hypothesis test for identifying whether one time series is a factor
and can provide relevant data for forecasting another time series (Li, 2020). This test can be run once on a
variety of variable combinations using the grangers causation matrix function. It attempts to run the Granger
Causality test on each pair of variables in the input dataset, in other words (Prabhakaran, 2022). The X and
Y variables are displayed in the output's columns and rows, respectively, along with the p-values from the
Granger causality test. P value can identify the significance of the correlation and causality. If the value of
p <0.05 (significance level), the X causes Y(Prabhakaran, 2022). Figure 4.6 shows the codes for granger
causality test. Grangercausalitytests function was used in Python to decide the number of lags applied for
TLCC. Then P values were calculated to decide the significance level for developed TLCC. If the values
of P were larger than 0.05 (significance level), the correlations would be removed from the groups of
correlations which were applied for establishing predicted models.

57

Figure 4.6 Granger Causality Test coding
4.2.2 Processing the developed Cross correlations by person
When processing the developed correlations, there are two parts needed to be concerned for selecting
reasonable correlations which can be used in the predicted model: how one signals leading another and the
value of P.
 TLCC & WTLCC - How one signals leading another
For TLCC and WTLCC, the location of the peak value indicates how one variable leads to another. For
example, there is a TLCC developed for S1 vs S2. If the peak value is the area of negative offsets, S1 is
leading the interaction, meaning S1 has more impact on S2. If the peak value is in the area of positive
offsets, S2 has more impact on S1. For this project, the correlations should be guaranteed that IEQ indicators
are leading the interaction with bio-signals so that the bio-signals can be used to predict IEQ parameters.
Take participant 1 as an example to explain the analysis for TLCC and WTLCC. Figure 4.7 shows the

58
WTLCC results for EDA vs. indoor temperature according to the data from participant 1.

Figure 4.7 WTLCC - EDA vs Temperature – Participant 1
It is apparent that the red area is larger than the blue area, which means the positive correlations is more
significant than the negative correlations. Therefore, there is a positive correlation between EDA and indoor
temperature. Additionally, TLCC graph can prove this result. Figure 4.8 shows the TLCC result.

Figure 4.8 TLCC - EDA vs Temperature – Participant 1

59
From the figure above, the peak value of the correlation coefficient is 0.647 and locates in the positive
offsets area which can verify the WTLCC results. Additionally, it is important to infer from the positive
offset that indoor temperature is leading the interaction with EDA data. Therefore, the correlation between
EDA and indoor temperature for participant 1 can be used in the Granger causality to identify if this
correlation can be used in the forecasting model.
However, there are also many results indicating false correlations which should be removed from Granger
Causality test. Figure 4.9 shows the WTLCC for heart rate vs indoor temperature according to the data from
participant 1.

Figure 4.9 WTLCC – Heart rate vs Temperature – Participant 1
Figure 4.9 shows that negative correlations are more significant than the positive correlations. Therefore,
there is a negative correlation between heart rate and indoor temperature. Figure 4.10 also verifies this
finding.

60

Figure 4.10 TLCC – Heart rate vs Temperature – Participant 1
However, the result from figure 4.10 indicates that although there is a negative correlation between heart
rate and indoor temperature, the peak value is in the area of negative offsets, meaning heart rate is leading
the interaction with indoor temperature. This kind of leading condition need to be removed from the next
step because it cannot be categorized into IEQ indicators impact human physiological responses.
Therefore, according these two kinds of conditions for TLCC and WTLCC graphs, some developed
correlations are remained to the next Granger causality test, but some of them are removed.

 Granger Causality Test – The value of P
From the demonstration in the 4.2.1, the value of P can decide the significance level of the correlations and
estimate if the correlations can be used in forecasting. If the value of P is from 0 to 0.05, the correlations
will be kept; if the value of P is from 0.05 to 1, the correlations should be removed. Table 4.1 shows the
results of participant 1. (Tem=indoor temperature, DiffPM=the difference of PM2.5 between 2 time

61
intervals, IAQ_E= IAQ evaluation, TC_E=Thermal comfort evaluation, HR=heart rate, STem=skin
temperature, SL=stress level)
p value Tem CO2 PM2.5 DiffPM RH Lighting Acoustics IAQ_E TC_E
EDA 0 0.597 0.0424 0.0249 0.687 0.0177 0.0174 0.0485 0.123
HR 0.0127 0.0787 0.096 0.122 0.079 0.138 0.234 0.486 0.0265
STem 0.0783 0.0493 0.0895 0.0382 0.0976 0.0937 0.126 0.976 0.257
SL 0.0262 0.278 0.0953 0.0674 0.0432 0.0282 0.01 0.004 0.0316
Table 4.1 p values results of participant 1
The values of p which are higher than 0.05 need to be removed because they are not significantly related
and there are not enough causality to do forecasting. Table 4.2 shows the process of removing the value of
p higher than 0.05. The green-colored cells are the parts not certificated into forecasting.
p value Tem CO2 PM2.5 DiffPM RH Lighting Acoustics IAQ_E TC_E
EDA 0 0.597 0.0424 0.0249 0.687 0.0177 0.0174 0.0485 0.123
HR 0.0127 0.0787 0.096 0.122 0.079 0.138 0.234 0.486 0.0265
STem 0.0783 0.0493 0.0895 0.0382 0.0976 0.0937 0.126 0.976 0.257
SL 0.0262 0.278 0.0953 0.0674 0.0432 0.0282 0.01 0.004 0.0316
Table 4.2 p values results removing p>0.05 of participant 1
Therefore, the remaining p values shows these correlations can be used to do the prediction model. Table
4.3 shows the final results of participant 1.
p value Tem CO2 PM2.5 DiffPM RH Lighting Acoustics IAQ_E TC_E
EDA 0

0.0424 0.0249

0.0177 0.0174 0.0485

HR 0.0127

0.0265
STem

0.0493

0.0382

SL 0.0262

0.0432 0.0282 0.01 0.004 0.0316
Table 4.3 p values final results of participant 1
 Combined the results from TLCC&WTLCC and Granger causality test
From the above explanation, the selected correlations need to meet two criteria: IEQ indicator is leading

62
bio signals, and the value of p<0.05. Table 4.4 combined these two results. The part of p>0.05 has already
been deleted and kept empty. Additionally, the green-colored cells mean the peak value of the cross
correlation coefficient is in the negative offsets, indicating that they need to be removed; the orange-colored
cells mean the peak value of the cross correlation coefficient is in the positive offsets or the middle, referring
to save to the forecasting model establishment. The red number is the p value, and the blue number is the
coefficient.
Coefficient
P value
Tem CO2 PM2.5 DiffPM RH Lighting Acoustics IAQ_E TC_E
EDA
-0.47
0

-0.55
0.0424
-0.07
0.0249

-0.453
0.0177
-0.18
0.0174
0.11
0.048-

HR
0.46
0.0127

0.339
0.0265
STem

0.319
0.0493

0.351
0.0382

SL
-0.45
0.0262

-0.38
0.0432
0.43
0.0282
0.268
0.01
0.4
0.004
0.392
0.0316
Table 4.4 Combined results from TLCC and p values of participant 1
The green-colored cells should be removed from the results because bio-signals are leading the interaction
with IEQ indicators. Table 4.5 shows the final results which will be used in establishing the prediction
models.
coefficient
P value
Tem CO2 PM2.5 DiffPM RH Lighting Acoustics IAQ_E TC_E
EDA
-0.47
0

-0.55
0.0424

-0.07
0.0249

-0.18
0.0174

0.11
0.048-

HR

0.339
0.0265

STem

SL
-0.45
0.0262

-0.38
0.0432

-0.38
0.0282

0.268
0.01

0.4
0.004

63
Table 4.5 Final selected correlations for forecasting of participant 1
Therefore, for participant 1, EDA can be used to predict 5 IEQ indicators: indoor temperature, PM2.5, the
difference of PM2.5 between 2 time intervals, acoustics level, IAQ evaluation; Heart rate can be used to
predict thermal comfort evaluation; Stress level can be used to predict indoor temperature, relative humidity,
lighting condition, acoustics level and IAQ evaluation; but skin temperature is not significantly correlated
with IEQ indicators’ prediction.
4.2.3 Processing the developed Cross correlations by gathering all participants’ data
Compared with the method applied in 4.2.1 which is analyzed by person, the approach for gathering all
participants’ data is almost same. The only discrepancy between them is the utilized dataset. The former
one used each participant’s data as dataset and repeated the process 14 times to complete it, but this section
will gather all the participants’ data into one excel sheet as a general dataset to do the cross correlation
analysis only once. Table 4.6 shows the process removing unsignificant p values. And table 4.7 shows the
results after combining TLCC and p values from Granger causality test. Similarly with 4.2.1, the green-
colored cells should be removed from the results because bio-signals are leading the interaction with IEQ
indicators. Table 4.8 is the final selected correlations for forecasting.
p value Tem CO2 PM2.5 DiffPM RH Lighting Acoustics IAQ_E TC_E
EDA 0.0123 0.044

4.65E-11

0.0045
HR 0.02 0.005

0.002 0.013

0.04
STem 0.04

0.02

SL 0.04

0.03
Table 4.6 p values final results based on combining all the data

64

coefficient
P value
Tem CO2 PM2.5 DiffPM RH Lighting Acoustics IAQ_E TC_E
EDA
-0.058
0.0123

-0.075
0.044

0.187
4.65E-11

-0.184
0.0045

HR
-0.046
0.002

-0.366
0.005

-0.301
0.02

-0.216
0.013

0.204
0.04

STem
0.034
0.04

0.061
0.02

SL
-0.122
0.04

0.246
0.03
Table 4.7 Combined results from TLCC and p values based on combining all the data

coefficient
P value
Tem CO2 PM2.5 DiffPM RH Lighting Acoustics IAQ_E TC_E
EDA
-0.058
0.0123

0.187
4.65E-11

-0.184
0.0045

HR
-0.046
0.002

-0.366
0.005

0.204
0.04

STem

0.061
0.02

SL
-0.122
0.04

0.246
0.03
Table 4.8 Final selected correlations for forecasting based on combining all the data
4.3 The impact of IEQ factors on Human Physiological responses indicators
4.3.1 Ranking significance level by person
According to the 4.2.2, all the data for 14 participants were processed by using TLCC, WTLCC and Granger
causality test in 14 tables. The correlations kept in the tables met with 2 criterial: IEQ indicators are leading
the correlation with bio-signals; the value of p is lower than 0.05, meaning a significant correlated

65
relationship. In order to rank the correlations, the number of each kind of correlation was counted among
14 tables. Table 4.7 shows the distribution of all the correlations.
Cases Tem CO2 PM2.5 DiffPM RH Lighting Acoustics IAQ_E TC_E
EDA 5 1 5 2 3 4 4 2 2
HR 2 1 2 1 4 3 3 1 4
STem 3 0 2 4 5 2 3 1 3
SL 1 1 2 2 4 3 3 2 1
Table 4.9 The number of significant correlation cases by persons
For example, when considering EDA VS indoor temperature, there are 5 volunteers’ data that have this
correlation which met the requirement in TLCC and Granger causality. However, there are only 1
volunteer’s data has the CO 2 VS EDA correlation. Therefore, it is apparent to conclude indoor temperature
has more impact on EDA than CO 2. Then, in order to show it more clearly, all the numbers summarized in
table 4.9 divided by 14, which is the percentage of all 14 participants. In this way, all the results are from 0
to 1 and a heatmap was established to show the significance of correlations by colors.

Figure 4.11 Heatmap showing the percentage of significant correlations cases based on personal data
From figure 4.11, it is clear that the darker the color, the more significant the correlations. Consequently,

66
there are three correlations showing the highest significance: indoor temperature with EDA, PM2.5 with
EDA, and relative humidity with skin temperature. However, CO 2 with skin temperature has no significant
correlation. Additionally, the averages of coefficients are calculated and a heatmap is also generated in
figure 4.12. The blue color indicates a negative correlation and the red color shows a positive correlation.

Figure 4.12 Average coefficient heatmap based on personal data
Additionally, a histogram was generated for illustrating each bio-signal’s significance with IEQ. Figure
4.13 shows the results.

Figure 4.13 Histogram for the number of significant correlation cases for each bio-signal based on
personal data
28
21
23
19
0
5
10
15
20
25
30
EDA Heart rate Skin temperature Stress level
The number of exsisting correlations for each biosignal by persons

67
From the figure above, EDA shows the highest correlation with IEQ indicators, but stress level has a
relatively lowe correlation with IEQ indicators. Similar histogram is generated for each IEQ indicators to
show the significance with human physiological responses. Figure 4.14 shows the results.

Figure 4.14 Histogram for the number of significant correlation cases for IEQ indicators based on
personal data
For IEQ indicators, relative humidity has the highest impact on bio-signals and CO 2 has the lowest impact
on bio-signals. Therefore, relative humidity needs to be paid more attention to improve the IEQ, but people
are not very sensitive to the change of CO 2.
4.3.2 Ranking significance level by genders
In order to rank the correlations by gender, the number of each kind of correlation was counted for each
gender group among 14 tables. There are 6 males and 8 females joining in the experiment. Table 4.10 shows
the number of selected correlations for females and Table 4.11 shows the number of selected correlations
for males.

68
The number of significant correlation cases for each correlation of female participants
Cases Tem CO2 PM2.5 DiffPM RH Lighting Acoustics IAQ_E TC_E
EDA 2 0 3 1 2 1 3 1 1
HR 2 0 1 0 2 2 1 1 3
STem 1 0 0 1 2 0 0 1 1
SL 1 0 0 0 1 2 2 1 1
Table 4.10 The number of significant correlation cases by gender - Females
The number of significant correlation cases for each correlation of male participants
Cases Tem CO2 PM2.5 DiffPM RH Lighting Acoustics IAQ_E TC_E
EDA 3 1 2 1 1 3 1 1 1
HR 0 1 1 1 2 1 2 0 1
STem 2 0 2 3 3 2 3 0 2
SL 0 1 2 2 3 1 1 1 0
Table 4.11 The number of significant correlation cases by gender – Males
Then, in order to show it more clearly, all the numbers summarized in table 4.8 divided by 8 which is the
percentage of all 8 female participants and the numbers summarized in table 4.9 divided by 6 which is the
percentage of all 6 male participants. In this way, all the results are from 0 to 1 and a heatmap was
established to show the significance of correlations by colors. Figure 4.15 and 4.16 shows the comparison
between females and males.

Figure 4.15 Heatmap showing the percentage of significant correlations cases by gender - Males

69
For males, indoor temperature and lighting conditions have a more significant correlation with EDA; PM 2.5
difference, relative humidity, and acoustics have a more significant correlation with skin temperature;
relative humidity impacts stress level more. However, there are no significant correlations between indoor
temperature, IAQ evaluation and heart rate; CO 2, IAQ evaluation and skin temperature; indoor temperature,
thermal comfort evaluation with stress level.

Figure 4.16 Heatmap showing the percentage of significant correlation cases by gender - Females
However, for females, there are only 3 correlations showing the more significant correlations: PM 2.5,
acoustics with EDA, and thermal comfort evaluation with heart rate. Additionally, there are 10 no
significant correlations, which is much higher than the male group.
Additionally, a histogram was generated for illustrating the percentage of each bio-signal’s significant
correlations with IEQ for females and males. Figure 4.17 shows the comparison.

70

Figure 4.17 The comparison between females and males for each bio-signals
For EDA and stress level, females are more easily influenced by IEQ indicators than males, but for heart
rate and skin temperature, males are more easily influenced by IEQ indicators than females. Figure 4.18
shows the comparison of IEQ indicators’ influence between females and males.

Figure 4.18 The comparison between females and males for each IEQ indicator
From the figure above, IEQ indicators impact males more than females when they are in the same
environment, except IAQ and thermal evaluation. That means, although males are easier to be impact by
0.19
0.00
0.13
0.06
0.22
0.16
0.19
0.13
0.19
0.21
0.13
0.29 0.29
0.38
0.29 0.29
0.08
0.17
0.00
0.05
0.10
0.15
0.20
0.25
0.30
0.35
0.40
Temperature CO2 PM2.5 DiffPM RH Lighting Acoustics IAQ
evaluation
Thermal
comfort
evaluation
The number of exsisting correlations for each IEQ indicator by gender
/Total number of female and male's correlations for each IEQ indicator
Female Male

71
IEQ indicators, female participants are more sensitive when considering IAQ and thermal conditions than
males.
4.3.3 Ranking significance level by experimental spaces
In order to rank the correlations by experimental spaces, the number of each kind of correlation was counted
for each space among 14 tables. There are 4 volunteers in DT office and 10 volunteers in USC studio joining
in the experiment. Table 4.12 shows the number of selected correlations DT office and Table 4.13 shows
the number of selected correlations for USC studio.
The number of significant correlation cases for each correlation of DT office participants
Cases Tem CO2 PM2.5 DiffPM RH Lighting Acoustics IAQ_E TC_E
EDA 1 0 0 0 1 1 0 0 0
HR 0 0 0 0 2 0 0 0 0
STem 0 0 0 0 0 0 0 0 1
SL 0 0 0 0 0 0 1 0 0
Table 4.12 The number of significant correlation cases by spaces – DT office
The number of significant correlation cases for each correlation of USC studio participants
Cases Tem CO2 PM2.5 DiffPM RH Lighting Acoustics IAQ_E TC_E
EDA 4 1 5 2 2 3 4 2 2
HR 2 1 2 1 2 3 3 1 4
STem 3 0 2 4 5 2 3 1 2
SL 1 1 2 2 4 3 2 2 1
Table 4.13 The number of significant correlation cases by spaces – USC Studio
Then, in order to show it more clearly, all the numbers summarized in table 4.10 divided by 4 which is the
percentage of all 4 participants in DT office and the numbers summarized in table 4.13 divided by 10 which
is the percentage of all 10 participants in USC studio. In this way, all the results are from 0 to 1 and a
heatmap was established to show the significance of correlations by colors. Figure 4.19 and 4.20 shows the

72
comparison between these two spaces.

Figure 4.19 Heatmap showing the significance of correlations summarized by spaces -DT office
For DT office, there are few significant cases from figure 4.19. That is because there were many
interruptions happened unexpectedly, such as having meeting, phone call or discussion with others. Only
relative humidity has more impact on volunteers’ heart rate. Acoustics shows a relatively correlated with
stress level, which is similar with thermal comfort evaluation VS skin temperature, lighting, relative
humidity, indoor temperature VS EDA.

73

Figure 4.20 Heatmap showing the significance of correlations summarized by spaces – USC studio
Comparing with DT office, USC studio shows more correlations. There is only one correlation does not
exist. All the other correlations show higher significance, especially for relative humidity VS skin
temperature and PM2.5 VS EDA. Therefore, because there is lack of correlations in DT office spaces
comparing with USC studio, the generated prediction model in chapter 5 should based on USC studio
dataset instead of DT office dataset.
Additionally, a histogram was generated for illustrating each bio-signal’s significance with IEQ for USC
studio. Figure 4.21 shows the result.

Figure 4.21 Histogram for the number of significant correlation cases of bio-signals based on USC studio
25
19
22
18
0
5
10
15
20
25
30
EDA Heart rate Skin temperature Stress level
The number of exsisting correlations for each biosignal in USC studio

74
data
From figure 4.21, EDA shows more significantly correlated with IEQ indicators. Heart rate shows less
correlated with IEQ indicators. However, there are not too many differences among all the bio-signals for
the correlations. Therefore, it is concluded that these four bio-signals all show significant correlations with
IEQ indicators.

Figure 4.22 Histogram for the number of significant correlation cases of IEQ indicators based on USC
studio data
Among all the IEQ indicators, relative humidity shows the most impact on bio-signals and CO2 shows the
less impact on bio-signals. This result is totally same with the result in 4.3.1 (by all persons), so it is
concluded that USC studio results can stand for the whole dataset and the influence of data in DT office
can be disregarded.
4.3.4 Ranking significance level by gathering all participants’ data
According to the 4.2.3, all the data for 14 participants were gathered together to process analysis by using
TLCC, WTLCC and Granger causality test in one general tables. The correlations kept in the tables met

75
with 2 criterial: IEQ indicators are leading the correlation with bio-signals; the value of p is lower than 0.05,
meaning a significant correlated relationship. In order to rank the correlations, p values are used here to
compare the significance of each correlation. Table 4.14 shows the p values and figure 4.23 shows the
heatmap.
P value Tem CO2 PM2.5 DiffPM RH Lighting Acoustics IAQ_E TC_E
EDA 0.0123 4.65E-11 0.0045
HR 0.02 0.005 0.04
STem 0.02
SL 0.04 0.03
Table 4.14 The significant correlation cases p values based on all the combined individual datasets
In figure 4.23, the smaller or closer the p-value is to 0, the greater the correlation. Therefore, the correlations
with lighter colors show more significant correlations.

Figure 4.23 Heatmap showing the percentage of significant correlation cases based on all the combined
individual datasets
From figure 4.23, it is clear that there are not too many significant correlations. Relative humidity has the
most significant impact on EDA. Additionally, the averages of coefficients are calculated and a heatmap is

76
also generated in figure 4.24. The blue color indicates negative correlation and red color shows positive
correlation.

Figure 4.24 Average coefficient heatmap based on all the combined individual datasets

4.4 Summary
To create the database, the data organization procedure is initially demonstrated in Chapter 4. Then, in order
to look into the significant correlations between human physiological response indicators and IEQ
indicators, data analysis techniques such as Time Lagged Cross Correlation (TLCC) & Windowed TLCC
analysis as well as Granger causality tests are presented.
When discussing the results of correlations based on individual datasets, there are three correlations
showing the highest significance: indoor temperature with EDA, PM2.5 with EDA, and relative humidity
with skin temperature. For all bio-signals, EDA shows the highest correlation with IEQ indicators, but stress
level has a relatively lower correlation with IEQ indicators. For IEQ indicators, relative humidity has the

77
highest impact on bio-signals and CO 2 has the lowest impact on bio-signals.
When discussing the significant correlations based on different genders data, for males, indoor temperature
and lighting conditions have a more significant correlation with EDA; PM 2.5 difference, relative humidity,
and acoustics have a more significant correlation with skin temperature; relative humidity impacts stress
level more. But for females, the significant correlations are much less than the males’ correlations.
Considering the bio-signals, females’ EDA and stress level are more easily influenced by IEQ indicators
than males, but males’ heart rate and skin temperature are more easily influenced by IEQ indicators than
females. Additionally, the results show although males are easier to be impacted by IEQ indicators, female
participants are more sensitive when considering IAQ and thermal conditions than males.
Considering different experiment spots, there are rare significant correlation in DT office environment.
Therefore, USC studio results stand for the whole dataset and the influence of data in DT office are
disregarded in the future prediction process.
When combined all participants’ data into one dataset to do the analysis, Relative humidity has the most
significant impact on EDA. And there are a smaller number of significant correlations than the correlations
analyzed individually.

78
CHAPTER 5: DISCUSSION OF THE PREDICTION MODELLING
Based on the description in chapter 4, IEQ indicators impact human physiological responses showing a lot
of significant correlations, which are approved as reasonable forecasting evidence. In this chapter, there are
two kinds of prediction models generated for future discussion. One is using 10 participants' data in USC
studio to develop 10 prediction models individually because the data from DT office shows rare significant
correlations (results from chapter 4). Another one establishes one prediction model based on gathering all
10 participants’ data into one dataset. At last, some comparisons are discussed for recommendations about
possible utilized scenarios and how to use these models.
5.1 Theory for establishing the prediction models
In order to generate the prediction models for time-series data, the Supervised learning method is applied
as the machine learning type and Random Forest is selected as the Supervised learning algorithm for both
personal data and general data.
5.1.1 Supervised learning
Supervised learning, commonly referred to as supervised machine learning, is a branch of artificial
intelligence and machine learning. It is characterized by the way it trains computers to accurately classify
data or predict outcomes using labeled datasets. The model modifies its weights as input data is fed into it
until the model has been properly fitted, which takes place as part of the cross validation process. For
classifying spam in a different folder from your email, supervised learning assists enterprises in finding
scaleable solutions to several real-world issues (What Is Supervised Learning? | IBM, n.d.).

79
In supervised learning, models are taught to produce the desired output using a training set. This training
dataset has both the right inputs and outputs, enabling the model to develop over time. The loss function
serves as a gauge for the algorithm's correctness, and iterations are made until the error is sufficiently
reduced (What Is Supervised Learning? | IBM, n.d.). There are mainly two types of problems when using
supervised learning: classification and regression.
 Classification can accurately allocate test data into specific categories by using algorithms. It identifies
particular entities in the dataset and makes an effort to determine how those things should be defined
or labeled.
 Regression is used to analyze the relationship between dependent and independent variables. It is
frequently used to produce estimates, including those for a company's sales revenue.
Supervised learning algorithms include mainly 7 types: Neural networks, Naive bayes, Linear regression,
Logistic regression, Support vector machines (SVM), K-nearest neighbor, and Random Forest. Among
these methods, Random Forest can be used for both classification and regression purposes, which is
appropriate for the time-series data. Therefore, Random Forest, as one of the common methods for data
prediction models, is utilized in this project.

5.1.2 Random Forest
A Random Forest Algorithm is a commonly used supervised machine learning technique that is utilized for
Classification and Regression issues in machine learning (Random Forest Algorithm, n.d.). To explain it,
the forest is the best example. A forest is made up of many different types of trees, and the more trees there

80
are, the more robust the forest will be. Similar to this, the accuracy and problem-solving capacity of a
Random Forest Algorithm increase with the number of trees in the algorithm. In order to increase the
dataset's predictive accuracy, a classifier called Random Forest uses many decision trees on different
subsets of the input data. It is based on the idea of ensemble learning, which is the practice of integrating
various classifiers to solve a challenging problem and enhance the model's performance (Random Forest
Algorithm, n.d.). Because of its excellent speed, scalability, and usability, the random forest technique (or
model) has greatly increased in prominence in machine learning (ML) applications during the past ten years.
It can tolerate redundant feature columns since it is adaptable and automatically gives feature priority scores.
It is resilient to overfitting and scales to huge datasets. The technique can simulate a nonlinear relationship
without the data having to be scaled (Tatsat et al., 2020). Figure 5.1 shows the algorithm used for
experimental data.

Figure 5.1 Random Forest prediction model algorithm example
The prediction models are generated for forecasting IAQ evaluation and thermal comfort evaluation based
on 4 bio-signals: heart rate, skin temperature, stress level and EDA, shown in the first line in this algorithm.
X is set as 4 bio-signals and Y is set as IAQ evaluation and thermal comfort evaluation, which are the

81
prediction target indicators. For both X and Y, data are separated into two groups for Random Forest: one
is for training and another one is for testing. ‘RandomForestClassifier’ function is used in Python for
generating prediction models and ‘accuracy_score’ function is utilized for deciding the accuracy of the
developed prediction model. During the process of establishing prediction models, the scale of testing data
is in the range of 30% to 40% of all data for the higher accuracies of models. For example, in figure 5.1,
the test size is 33% of personal data.

5.2 Prediction model results and accuracy
For the prediction models, there are mainly two parts: one is based on individual data to establish 10
prediction models for 10 participants; another one is based on general data which is the dataset after
gathering all 10 participants' data into one general sheet.
5.2.1 Individual prediction models
There are 10 participants’ data used for generating prediction models, who are all in the USC studio
experiment. After doing the Random Forest prediction model, the predicted data are outputted with actual
data to make a comparison at first. Take participant 1 as an example. Figure 5.2 shows the comparison
between IAQ evaluation prediction data and IAQ evaluation actual data. Figure 5.3 shows the comparison
between thermal comfort evaluation prediction data and thermal comfort evaluation actual data.

82

Figure 5.2 The difference between IAQ evaluation prediction data and actual data-Participant 1
From Figure 5.2, the hollow circle in orange color is the prediction data and the solid circle in blue color is
the actual data. In total, there are 44 sets of data for testing the accuracy. There are only 6 sets of data with
± 1 difference between prediction and actual data. The rest of them are all the same.

Figure 5.3 The difference between Thermal Comfort evaluation prediction and actual data-Participant 1
1
1
1
-1
1
1
-3.5
-3
-2.5
-2
-1.5
-1
-0.5
0
0 5 10 15 20 25 30 35 40 45 50
IAQ evaluation
The number of Test dataset
The Difference between Predicted and actual data for IAQ evaluation-Participant 1
IAQ Data IAQ Prediction

83
Similar to IAQ evaluation, there are 44 test datasets for prediction. Comparing with IAQ evaluation, thermal
comfort prediction data have lower accuracy because the range of difference is ± 2 and there are more cases
showing the discrepancy.
Then the accuracy of each participant’s prediction model is calculated with the algorithm. Figure 5.4 shows
all accuracies of 10 prediction models by ranking. Figure 5.4 shows the accuracies of IAQ evaluation
prediction models and Figure 5.5 shows the accuracies of thermal comfort evaluation prediction models.

Figure 5.4 The accuracies of IAQ evaluation prediction models

84

Figure 5.5 The accuracies of Thermal Comfort evaluation prediction models
From figure 5.4 and 5.5, the average of individual prediction models for IAQ evaluation is 84.76% and the
average of individual prediction models for Thermal Comfort evaluation is 70.5%. However, because IAQ
and thermal comfort conditions were hard for some participants to evaluation, some prediction models have
lower accuracy, which is acceptable.
Besides the calculation of the average, there is a comparison between females and males for IAQ and
thermal comfort evaluations’ prediction accuracies by using 2- sample T-test. A 2-sample t-test can
construct a confidence interval for the mean difference between females and males. Figure 5.6 shows the
T-test for IAQ evaluation prediction accuracies and Figure 5.7 shows the T-test for the Thermals Comfort
evaluation prediction accuracies.

85

Figure 5.6 T-test for IAQ evaluation prediction accuracies by genders(p=0.146)
For the IAQ prediction models, although the averages of accuracies between females and males have 10%
discrepancy, the range of accuracies has a lot of overlap, which means the results of these two groups have
many similarities and the discrepancy is not high enough to conclude a significant impact.
-
Figure 5.6 T-test for Thermal Comfort evaluation prediction accuracies by genders (p=0.473)
Similarly, the data range of the thermal comfort T-test shows even more overlaps between females and
males. Therefore, it is concluded that there is no effect of gender on IAQ and thermal comfort prediction

86
accuracies.

5.2.2 General prediction model
The general prediction model is generated based on gathering all participants’ data into one dataset. The
same algorithm for Random Forest is used to establish a prediction model. Because of the larger scale of
the dataset, there are 605 sets of data, which is 33% of all data. Figure 5.7 shows the comparison between
the average of individual prediction models’ accuracies and the general model accuracy for IAQ evaluation.
Figure 5.8 shows the comparison between the average of individual prediction models’ accuracies and the
general model accuracy for thermal comfort evaluation.

Figure 5.7 IAQ evaluation prediction accuracies comparison between individual prediction average
accuracy and general prediction accuracy
84.76%
69.70%
0.00%
10.00%
20.00%
30.00%
40.00%
50.00%
60.00%
70.00%
80.00%
90.00%
100.00%
IAQ evaluation predcition average
accuracy in individual model
IAQ evaluation predcition accuracy in
general model
IAQ evaluation accuracy comparison
IAQ evaluation
predcition
average
accuracy in
individual
model
IAQ evaluation
predcition
accuracy in
general model

87

Figure 5.8 Thermal Comfort evaluation prediction accuracies comparison between individual prediction
average accuracy and general prediction accuracy
Both the thermal comfort prediction accuracy and the IAQ prediction accuracy are lower than the average
individual prediction accuracy, which means the prediction accuracy will be better when considered by
person instead of by the group.
5.3 Comparison and discussion of the results
5.3.1 Comparison between the individual prediction model and general prediction model
Figure 5.9 and 5.10 shows the comparison of all the accuracies output from Random Forest including
individual models and the general model for IAQ evaluation and Thermal Comfort evaluation predictions.

88

Figure 5.9 All IAQ evaluation prediction accuracies comparison
For IAQ evaluation prediction models, the average individual prediction accuracy (84.76%) is higher than
the general prediction accuracy (69.7%). Except for only one participant’s prediction model, all individual
prediction models have higher accuracy than general model prediction accuracy. Therefore, the individual
prediction model is recommended for future usage because of its higher accuracy for Indoor Air Quality
evaluation.

89

Figure 5.10 All Thermal Comfort evaluation prediction accuracies comparison
For Thermal Comfort evaluation prediction models, the average individual prediction accuracy (70.5%) is
higher than the general prediction accuracy (64.3%). Except for three participants’ prediction model, all
individual prediction models have higher accuracy than general model prediction accuracy. Therefore, the
individual prediction model is recommended for future usage because of its higher accuracy for Thermal
Comfort evaluation. However, compared with IAQ evaluation prediction model, the accuracies of Thermal
Comfort prediction model are lower, meaning more difficulties when predicting thermal comfort condition
than IAQ conditions.
In general, the individual prediction models’ accuracy for both IAQ and thermal comfort is better than the
general model’s accuracy. Therefore, it is recommended the establishment or application of the predictive
model should be preferred in a personal independent space or a space with a small number of occupants for
a better performance of prediction.

90
5.3.2 Discussion of participants’ weight, height, and BMI with individual prediction
Because individual prediction models have higher accuracy, the potential reasons are discussed about the
participants’ body data. There are two indicators: weight and height. In order to combine them together,
body mass index is used here. Body Mass Index can estimate the amount of body fat by using weight in
kilograms (kg) divided by the square of height in meters (m
2
).
Participants’ weights are recorded in Tables 5.1 and 5.2. Table 5.1 displays the participants' height, weight,
and BMI organized according to the predictive accuracy of the IAQ prediction models based on individual
data. And table 5.2 displays the same indexes organized according to the predictive accuracy of the Thermal
Comfort prediction models based on individual data.

IAQ prediction
accuracy rank from
high to low
IAQ prediction
accuracy
Weight
(lbs)
Average
weight (lbs)
Height
(ft)
Average
Height (ft)
BMI Average
BMI
S-1 100.0% 180

168.2
5.97

6.0
24.8

23.3
S-2 100.0% 163 6.59 18.3
S-3 93.8% 220 6.14 28.6
S-4 93.3% 138 5.31 24.0
S-5 84.8% 140 5.74 20.9
Average 84.7% 160 5.77 23.6
S-6 83.3% 195

151.48
5.74

5.6
29.1

23.6
S-7 79.7% 138 5.48 22.6
S-8 75.4% 180 6.07 24.0
S-9 74.0% 125 5.31 21.7
S-10 63.3% 119 5.31 20.6
Table 5.1 Participants' height, weight, and BMI organized according to IAQ individual prediction accuracy

91
Thermal Comfort
prediction accuracy
rank from high to low
Thermal Comfort
prediction
accuracy
Weight
(lbs)
Average
weight
(lbs)
Height
(ft)
Average
Height
(ft)
BMI Average
BMI
S-8 93.10% 180

163.8
6.07

5.9
24.0

23.0
S-2 86.40% 163 6.59 18.3
S-7 83.80% 138 5.48 22.6
S-3 81.20% 220 6.14 28.6
S-5 77.30% 140 5.74 20.9
S-9 74.00% 125 5.31 21.7
S-1 72.00% 180 5.97 24.8
Average 70.50% 160 5.77 23.6
S10 50.30% 119
150.7
5.31
5.5
20.6
24.5 S-4 44.70% 138 5.31 24.0
S-6 42.20% 195 5.74 29.1
Table 5.2 Participants' height, weight, and BMI organized according to Thermal Comfort individual
prediction accuracy
The yellow section in the table represents the body data of models whose predictive accuracy is above
average, while the blue section represents the body data of models whose predictive accuracy is below
average. In addition, the weight, height, and BMI of the volunteers in these two sections have also been
averaged separately.
For table 5.1 with IAQ accuracy, it is clear that the average weight and height of individuals above the
average predictive accuracy are higher than the overall average, and the average weight and height of
individuals below the average predictive accuracy are lower than the overall average. When it comes to
BMI, the situation is reversed. The average BMI of individuals above the average predictive accuracy is
lower than the overall average BMI, while the average BMI of individuals below the average predictive
accuracy is higher than the overall average BMI.
Figure 5.11 shows the participants’ BMI and the average BMI in table 5.1. Figure 5.12 shows the

92
participants’ BMI and the average BMI in table 5.2.

Figure 5.11 Participants’ BMI and average BMI according to IAQ prediction accuracy rank

Figure 5.12 Participants’ BMI and average BMI according to Thermal Comfort prediction accuracy rank
Based on the chart, the analysis indicates that individuals with lower BMI are more likely to make accurate
IAQ predictions. This relationship is also observed in the context of thermal comfort, where a lower average

93
BMI of subjects is associated with increased predictive accuracy. Therefore, based on the current findings,
one can infer that individuals with lower BMI are more likely to accurately predict both IAQ and thermal
comfort.

5.4 Summary
In chapter 5, the supervised machine learning method is used for establishing prediction models. Random
Forest, as one of the most common and flexible supervised learning algorithms, is applied to develop the
models and output the accuracy. There are two categories of prediction models: one is based on 10
participants' data to establish 10 prediction models individually; another one is based on one dataset
combining all 10 participants’ data together to develop a general prediction model. IAQ evaluation and
thermal comfort evaluation are predicted separately based on 4 human physiological responses. The results
show that the individual prediction models’ accuracy for both IAQ and thermal comfort is better than the
general model’s accuracy. Meanwhile, the prediction accuracy of IAQ evaluation is higher than the
accuracy of thermal comfort prediction accuracy.
In the final discussion part, BMI has a negative impact on IAQ and thermal comfort. Therefore, the lower
amount of body fat is more likely to lead more accurate prediction of IAQ and thermal comfort.

94
CHAPTER 6: CONCLUSIONS AND FUTURE WORK
6.1 Conclusion
This research focused on the correlations between indoor environmental quality (IEQ) and human
physiological responses in the commercial area, with the aim of developing predicted models for an
automatic ventilation system based on physiological responses. The experiments took place in two spaces.
The first office had 4 volunteers (3 females and 1 male) and the second office had 10 volunteers (5 males
and 5 females) between the ages of 22 and 35, all healthy with no history of asthma. Each volunteer spent
1-3 days working or studying in the office for 4-6 hours each day. Sensors were installed in the middle of
each space to collect environmental data which was uploaded to a cloud platform. Outdoor PM2.5 sensors
were also installed since the offices were closed environments. Each participant wore two smartwatches to
collect bio-signals such as heart rate, skin temperature, EDA, and stress levels. Two surveys were filled out
every two hours to assess indoor thermal comfort and indoor air quality satisfaction. Data were collected
every 10 minutes except for the illuminance sensor which was tested at 10 am, 1 pm, and 5 pm since the
offices were usually artificially lit throughout the day.
After data collection, these data were organized into the same 10-minute time interval, which was analyzed
using various techniques, including Time Lagged Cross Correlation (TLCC) and Granger causality tests.
These two methods identified the significant correlations among these indicators and proved the capability
of forecasting. The results showed that there were significant correlations between certain IEQ indicators
and physiological responses, with indoor temperature, PM2.5, and relative humidity having the most

95
significant impact on EDA, skin temperature, and stress level, respectively, with a p-value lower than 0.05.
The correlations also showed differences between males and females, with males being more impacted by
IEQ indicators but females being more sensitive to indoor air quality and thermal comfort evaluation.
Additionally, when comparing the results from two spaces, the dataset in DT office was not capable of
forecasting because of less significant correlation cases compared with the dataset in USC studio.
Based on the correlation analysis, this study developed prediction models for IAQ and thermal comfort
evaluation using a supervised machine learning method, specifically Random Forest. The models showed
that individual prediction models based on 10 participants' data had better accuracy than a general model
based on a combined dataset. The study found that, on average, the individual prediction models for Indoor
Air Quality (IAQ) evaluation had a higher accuracy (84.76%) than the general prediction model (69.7%).
All but one participant's individual prediction model had higher accuracy than the general model, indicating
that the individual model was recommended for future use. Similarly, for Thermal Comfort evaluation
prediction models, the average individual prediction accuracy (70.5%) was higher than the general
prediction accuracy (64.3%), with all but three participants' individual prediction models having higher
accuracy than the general model. Therefore, the individual prediction model was also recommended for
future usage in Thermal Comfort evaluation. However, it should be noted that the accuracies of the Thermal
Comfort prediction models were lower compared to the IAQ prediction models, indicating more difficulty
in predicting thermal comfort conditions than IAQ conditions. Additionally, participants’ BMI had a
negative impact on prediction accuracy. The analysis showed the average BMI of people with predicted
accuracy higher than average was 23.3, which was lower than 23.6 (the average BMI of people with

96
predicted accuracy lower than average). Therefore, individuals with lower BMI were more likely to
accurately predict both IAQ and thermal comfort.
Overall, the study provided insights into the relationship between IEQ and human physiological responses,
as well as the potential for using physiological responses as indicators for improving ventilation control
systems. The IAQ and thermal comfort standard should relay on the female groups’ evaluation to improve
the indoor comfort. Additionally, because the people with lower BMI are more likely to have a more
accurate predicted model, the general model can be based on lower BMI people’s group to make a higher
accuracy.
6.2 Future Work
Although the experiment yielded some reasonable results about the data correlations, there are many data
discontinuities and inaccuracies that occurred during the analysis process which can impact the developed
correlation and validated models. Therefore, some improvements and future works are suggested to achieve
more comprehensive and accurate results.
6.2.1 Limitation of the current workflow
There are mainly three limitations in the experiment that should be considered for improving.
The first component is about experimental devices: smartwatches, and IEQ sensors. For smartwatches,
during the data collection and organization, there are many data gaps occurred because of smartwatches’
technical limitations. For example, when the smartwatches are loose or do not fit snugly on the tester's arm,
the data can’t be collected so that the data is empty for that period. Additionally, the Empatica embrace 2

97
smartwatches have some unexpected connection problems which lead to data failures. Occasionally, if the
participants are a little away from the connected phones or forget to keep their phones on, the app connected
with Empatica embrace 2 smartwatches will stop recording and the data stored on the data cloud is possible
to disappear. Therefore, some improvements in smartwatches can be done to enhance accuracy and data
reliability. For IEQ sensors, the HOBO sensor is the most convenient device for data accessibility and
accuracy. However, lighting sensors and PM 2.5 sensors are not convenient enough for collecting data.
Lighting sensors can’t automatically store data so researchers should manually record at specific time points.
This manual test leads to discontinuous data and inaccuracy. PM 2.5 sensors are required to connect with a
stable WIFI to collect data, but some WIFI unstable situations can’t be controlled so some data are missed.
Additionally, the dimension of PM 2.5 tests is not large enough to cover the lower ranges when indoor PM 2.5
conditions are good. Furthermore, the installation of PM 2.5 sensors is difficult, which challenges the
experiment process. Therefore, more precise PM 2.5 sensors are recommended for future work.
The second part is about the status of involving volunteers. Processed data is unstable when volunteers are
sleeping or walking around and this problem impacted the results. Therefore, researchers should request
volunteers to keep working, studying or doing desktop work instead of walking, playing or sleeping.
The third limitation is the method of developing correlation analysis and predicted model is not
comprehensive enough. In this experiment, only TLCC and Granger Causality were used for correlation
analysis and Random Forest used for establishing predicted models. A better analysis can be pursued by
applying more correlation analysis methods and Machine learning algorithms to validate the accuracy of
the correlations and predicted models. Multiple analysis methods are possible to validate results and achieve

98
a persuasive dataset. Additionally, after developing predicted models, a repeated experiment can be done
for one or two volunteers to verify the model's accuracy and approve the method's availability.
The final limitation is the overfitting problem when doing supervised Machine Learning. When a machine
learning model makes correct predictions for training data but not for new data, this is a bad machine
learning behavior known as overfitting. The supervised machine learning models were only trained on
known data sets when used to make predictions. The model then tried to forecast results for new data sets
based on this information. An overfit model cannot perform well for all sorts of new data and can provide
inaccurate predictions. In this research, because the dataset is not large enough, overfitting problem could
happen and lead to inaccurate results.

6.2.2 Recommendations for future practice
About the volunteers of the experiments, there are three parts recommended for future practice. First of all,
the number of volunteers can be increased. The results in this research are based on a 200-time points
dataset for each participant in only two environments, so it is hard to ensure the results can be adapted to
more participants. After enlarging the scale of the dataset, the results will be more accurate and universal
for common people. Secondly, the number of experimental spots can be increased. This research only
carries on experiments in the DT office and USC MBS studio. Apparently, the number of experimental
spots is not enough to fit more scenarios. In the future, more spaces and scenarios can be considered to add
to experiments. There are 4 scenarios listed below for future applications and studies considering the
prediction model plays better in a room with a single occupant or a small number of occupants.

99
 Health and well-being buildings – Hospitals
In the hospital, there are many single or multiple-patient rooms, where a small number of patients need
to stay and live for a while. Figure 6.1 shows the example picture of the patient room. Especially in
Intensive Care Units, some patients usually can’t express their commands by themselves, but the
indoor environment quality should be adjusted accurately for improving the patient’s healing process.
In that case, the prediction model technology is recommended for the HVAC system to achieve
automatic control. Firstly, human bio-signals are easier to collect in the patient room because there are
many health monitoring systems for recording patients’ basic bio-signals, such as heart rate, blood
pressure, and skin temperature. Secondly, the patients need to stay in one patient room for a long time.
Therefore, the prediction model can be established in several days by using the monitored human bio-
signals data. Finally, the prediction models can be automatically kept testing and validating when
patients are occupied to improve the accuracy. It is desired that this prediction system can help those
patients, who are unconscious or hard to express self-command, to adapt to the indoor environment.

Figure 6.1 Single-patient room (Kontakt, 2021)

100
 Education facilities – Kindergarten or Elementary school
Kindergarten or elementary class is commonly in a small number of children. Additionally, the kids
are one of the most vulnerable group of people because they are not old enough to adjust indoor
environment well. Considering kids easily get sick because of bad IEQ, indoor quality evaluation
predicted model is a convenient way to enhance the indoor environmental quality and reduce the risk
of illness. Multiple diffusers can be used in one room for different space functions. For example, the
entertainment space has a diffuser to make kids adapt to the active environment. Meanwhile, the
reading space can install another diffuser to improve the stationary environmental quality. Figure 6.2
shows the environment of suggested education facility environment.

Figure 6.2 Predicted HVAC system in suggested Kindergarten environment (Merrill, n.d.)
 Commercial spaces – Individual office or small meeting room
In order to improve productivity and employees’ health conditions, individual office or small meeting
rooms are recommended to apply this predicted model technology in HVAC. Sometimes, the long

101
meeting time makes occupants dizzy and hard breathing. That is usually because the indoor air quality
is poor, and the indoor air does not stagnant well in a closed environment. Therefore, prediction mode
is recommended for convenient HVAC control and better indoor adaption. Figure 6.3 shows an
example in the commercial building.

Figure 6.3 Predicted HVAC system in the suggested meeting room (InsideOut Studios, n.d.)
Thirdly, the group of participants is not enough to cover most of the ages of people. It is recommended to
have more participants whose ages range is wider. Especially for the elder and younger groups, these groups
of people may show different results and the results will be valuable for future investigations.
Regarding the data analysis process, it is recommended to make a functional excel to show all the
correlation analysis and predicted model results because the results are separated in different files and are
hard to find directly when some data are needed. Additionally, more data-analysis software can be used to
validate the results and develop a more persuasive model. Furthermore, artificial intelligence can be applied
for the prediction model to achieve scientific visualization about the results.

102
Fourthly, participants should wear smartwatches for one or two days before the experiment to adapt
smartwatches. In this way, the collected data will be more stable and accurate.
Finally, after the same experiments are done in different kinds of spaces, the general models can be
generated based on low BMI groups. This general model can be used as a referred model to improve the
accuracy of predicted models for high BMI group.

103
BIBLIOGRAPHY
Active sampler for microbiological air sampling “Microflow alfa” | Aquaria Srl. (n.d.). Retrieved
September 13, 2022, from http://www.aquariasrl.com/en/campionatore-attivo-per-il-campionamento-
microbiologico-dellaria-microflow-alfa/

Adamou, M., Fullen, T., & Jones, S. L. (2020). EEG for Diagnosis of Adult ADHD: A Systematic Review
With Narrative Analysis. Frontiers in Psychiatry, 11. https://doi.org/10.3389/FPSYT.2020.00871/FULL

Aerocet 831 Handheld Aerosol Mass Monitor - Met One Instruments. (n.d.). Retrieved September 13, 2022,
from https://metone.com/products/aerocet-831-handheld-particle-counter/

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

ASHRAE. (2016). ASHRAE Guideline 10-2016 - Interactions Affecting the Achievement of Acceptable
Indoor Environments. https://webstore.ansi.org/Standards/ASHRAE/ASHRAEGuideline102016

Ayoub Meo, S., Adnan Abukhalaf, A., Sami, W., & Hoang, T. D. (2021). Effect of environmental pollution
PM2.5, carbon monoxide, and ozone on the incidence and mortality due to SARS-CoV-2 infection in
London, United Kingdom. Journal of King Saud University - Science, 33(3), 101373.
https://doi.org/10.1016/J.JKSUS.2021.101373

Bandgar, S. (2021). Support Vector Machine.. Support vector machines (SVMs) are… | by Swapnil
Bandgar | Nerd For Tech | Medium. Nerd For Tech. https://medium.com/nerd-for-tech/support-vector-
machine-92fa3c57d33b

Benabed, A., & Boulbair, A. (2022a). Numerical analysis of thermal comfort and air freshness generated
by a multi-cone diffuser with and without lobed inserts. Journal of Building Engineering, 54, 104632.
https://doi.org/10.1016/J.JOBE.2022.104632

Benabed, A., & Boulbair, A. (2022b). Numerical analysis of thermal comfort and air freshness generated
by a multi-cone diffuser with and without lobed inserts. Journal of Building Engineering, 54, 104632.
https://doi.org/10.1016/J.JOBE.2022.104632

Best Smart Vent: Room by room climate control. (n.d.). Retrieved November 17, 2022, from
https://thesmartcave.com/best-smart-vent/

104
Bryn Farnsworth. (2019). What is EDA? And how does it work? https://imotions.com/blog/eda/

Cain, R., Jennings, P., & Poxon, J. (2013). The development and application of the emotional dimensions
of a soundscape. Applied Acoustics, 74(2), 232–239. https://doi.org/10.1016/J.APACOUST.2011.11.006

Cony Renaud Salis, L., Abadie, M., Wargocki, P., & Rode, C. (2017). Towards the definition of indicators
for assessment of indoor air quality and energy performance in low-energy residential buildings. Energy
and Buildings, 152, 492–502. https://doi.org/10.1016/J.ENBUILD.2017.07.054

Dastbaz, M., Halpin, E., & Wright, S. (2013). Emerging Technologies and the Human Rights Challenge of
Rapidly Expanding State Surveillance Capacities. Strategic Intelligence Management: National Security
Imperatives and Information and Communications Technologies, 108–118. https://doi.org/10.1016/B978-
0-12-407191-9.00010-7

Data logger for humidity, temperature, air pressure and CO2 in HVAC applications. (n.d.). Retrieved
September 13, 2022, from https://www.epluse.com/products/co2-measurement/co2-data-logger/humlog20/
Dedicated Outside Air System DOAS - MEP Academy. (2022). https://mepacademy.com/dedicated-
outside-air-system-doas/

Deng, M., Wang, X., & Menassa, C. C. (2021). Measurement and prediction of work engagement under
different indoor lighting conditions using physiological sensing. Building and Environment, 203, 108098.
https://doi.org/10.1016/J.BUILDENV.2021.108098

Draghici, A. E., & Taylor, J. A. (2016). The physiological basis and measurement of heart rate variability
in humans. Journal of Physiological Anthropology, 35(1). https://doi.org/10.1186/S40101-016-0113-7

Du, H., Lian, Z., Lai, D., Duanmu, L., Zhai, Y., Cao, B., Zhang, Y., Zhou, X., Wang, Z., Zhang, X., & Hou,
Z. (2022a). Evaluation of the accuracy of PMV and its several revised models using the Chinese thermal
comfort Database. Energy and Buildings, 271, 112334. https://doi.org/10.1016/J.ENBUILD.2022.112334

Du, H., Lian, Z., Lai, D., Duanmu, L., Zhai, Y., Cao, B., Zhang, Y., Zhou, X., Wang, Z., Zhang, X., & Hou,
Z. (2022b). Evaluation of the accuracy of PMV and its several revised models using the Chinese thermal
comfort Database. Energy and Buildings, 271, 112334. https://doi.org/10.1016/J.ENBUILD.2022.112334
Durier, F., Carrié , F. R., & Sherman, F. M. (2018). V entilation I nformation P aper n° 38 What is smart
ventilation? www.inive.org

Electroencephalogram (EEG) | Johns Hopkins Medicine. (n.d.). Retrieved September 8, 2022, from
https://www.hopkinsmedicine.org/health/treatment-tests-and-therapies/electroencephalogram-eeg
Electroencephalogram (EEG) - NHS. (n.d.). Retrieved September 8, 2022, from
https://www.nhs.uk/conditions/electroencephalogram/

105
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

Guenther, S. (2021). What Is PMV? What Is PPD? Basics of Thermal Comfort | SimScale. SIMSCALE
BLOG. https://www.simscale.com/blog/what-is-pmv-ppd/

Guillard, E. (2021, May 21). How to Measure the Indoor Air Quality of your Buildings - Dexma.
https://www.dexma.com/blog-en/how-to-measure-the-indoor-air-quality-of-your-buildings/

HAYES, A. (2022, June 12). What Is a Time Series and How Is It Used to Analyze Data?
https://www.investopedia.com/terms/t/timeseries.asp

Higuera-Trujillo, J. L., Ló pez-Tarruella Maldonado, J., & Llinares Millá n, C. (2017). Psychological and
physiological human responses to simulated and real environments: A comparison between Photographs,
360° Panoramas, and Virtual Reality. Applied Ergonomics, 65, 398–409.
https://doi.org/10.1016/J.APERGO.2017.05.006

How Important Is Indoor Air Quality To Your Health? - Air Control Guy. (n.d.). Retrieved October 21,
2022, from https://aircontrolguy.com/how-important-is-indoor-air-quality-to-your-health/

How to Measure Light Intensity: Understanding & Using a Lux Meter. (n.d.). Retrieved September 14,
2022, from https://bioslighting.com/how-to-measure-light-intensity/architectural-lighting/

Hwang, T., & Jeong, T. K. (2010). Effects of Indoor Lighting on Occupants’ Visual Comfort and Eye
Health in a Green Building. Https://Doi.Org/10.1177/1420326X10392017, 20(1), 75–90.
https://doi.org/10.1177/1420326X10392017

Indoor Air and Coronavirus (COVID-19) | US EPA. (n.d.). Retrieved January 31, 2023, from
https://www.epa.gov/coronavirus/indoor-air-and-coronavirus-covid-19

InsideOut Studios. (n.d.). Retrieved March 8, 2023, from
https://www.insideoutstudios.com/RICK%20PETERS/Architecture/

Jin Cheong. (2019, May 13). Four ways to quantify synchrony between time series data | by Jin Cheong,
PhD | Towards Data Science. https://towardsdatascience.com/four-ways-to-quantify-synchrony-between-
time-series-data-b99136c4a9c9

Joshi, S. (2008). The sick building syndrome. Indian Journal of Occupational and Environmental Medicine,
12(2), 61. https://doi.org/10.4103/0019-5278.43262

106
Juan, Y. K., & Chen, Y. (2022). The influence of indoor environmental factors on learning: An experiment
combining physiological and psychological measurements. Building and Environment, 221, 109299.
https://doi.org/10.1016/J.BUILDENV.2022.109299

Kalita, D. (2022). An Overview of Deep Belief Network (DBN) in Deep Learning.
https://www.analyticsvidhya.com/blog/2022/03/an-overview-of-deep-belief-network-dbn-in-deep-
learning/

Kang, S., Mak, C. M., Ou, D., & Zhang, Y. (2022). An investigation of acoustic environments in large and
medium-sized open-plan offices in China. Applied Acoustics, 186, 108447.
https://doi.org/10.1016/J.APACOUST.2021.108447

Kim, J., Hong, T., Jeong, J., Koo, C., & Kong, M. (2017). An integrated psychological response score of
the occupants based on their activities and the indoor environmental quality condition changes. Building
and Environment, 123, 66–77. https://doi.org/10.1016/J.BUILDENV.2017.06.046

Kim, J., Kong, M., Hong, T., Jeong, K., & Lee, M. (2018). Physiological response of building occupants
based on their activity and the indoor environmental quality condition changes. Building and Environment,
145, 96–103. https://doi.org/10.1016/J.BUILDENV.2018.09.018

Kim, J. L., Elfman, L., Mi, Y., Wieslander, G., Smedje, G., & Norbä ck, D. (2007). Indoor molds, bacteria,
microbial volatile organic compounds and plasticizers in schools – associations with asthma and respiratory
symptoms in pupils. Indoor Air, 17(2), 153 –163. https://doi.org/10.1111/J.1600-0668.2006.00466.X

Kontakt. (2021, May 5). Patient Room Monitoring System. Environmental Monitoring | Kontakt.io.
https://kontakt.io/blog/patient-room-monitoring-managing-the-environment-to-impact-clinical-outcomes/

Kurama, V. (2019). Introduction to Bagging and Ensemble Methods | Paperspace Blog.
https://blog.paperspace.com/bagging-ensemble-methods/

Lee, J.-H., Moon, J. W., & Kim, S. (2014). Analysis of Occupants’ Visual Perception to Refine Indoor
Lighting Environment for Office Tasks. Energies 2014, Vol. 7, Pages 4116-4139, 7(7), 4116–4139.
https://doi.org/10.3390/EN7074116

Li, S. (2020). A Quick Introduction On Granger Causality Testing For Time Series Analysis | by Susan Li
| Towards Data Science. https://towardsdatascience.com/a-quick-introduction-on-granger-causality-
testing-for-time-series-analysis-7113dc9420d2

MacGill, M. (2021). Heart rate: What is a normal heart rate?
https://www.medicalnewstoday.com/articles/235710#normal-resting-heart-rate

107
Madureira, J., Paciê ncia, I., Ramos, E., Barros, H., Pereira, C., Teixeira, J. P., & de Oliveira Fernandes, E.
(2015). Children’s Health and Indoor Air Quality in Primary Schools and Homes in Portugal—Study
Design. Https://Doi.Org/10.1080/15287394.2015.1048926, 78(13–14), 915–930.
https://doi.org/10.1080/15287394.2015.1048926

Maintaining thermal comfort in indoor work environments | SafeWork NSW. (n.d.). Retrieved August 31,
2022, from https://www.safework.nsw.gov.au/resource-library/heat-and-environment/maintaining-
thermal-comfort-in-indoor-work-environments

Mansi, S. A., Pigliautile, I., Arnesano, M., & Pisello, A. L. (2022). A novel methodology for human thermal
comfort decoding via physiological signals measurement and analysis. Building and Environment, 222,
109385. https://doi.org/10.1016/J.BUILDENV.2022.109385

Melikov, A., Grø nbaek, H., & Nielsen, J. B. (n.d.). Personal Ventilation: from research to practical use.
Merrill, S. (n.d.). 7 Outstanding K–8 Flexible Classrooms | Edutopia. 2017. Retrieved March 8, 2023, from
https://www.edutopia.org/article/7-outstanding-k-8-flexible-classrooms/

Mod, S. (n.d.). Wearable Technology All Set To Reform The World Of Biometrics. Retrieved October 21,
2022, from https://www.engineersgarage.com/wearable-technology-all-set-to-reform-the-world-of-
biometrics/

Nguyen, O. (2016). What is a Dedicated Outdoor Air System? - Refrigeration School, Inc. (RSI).
https://www.rsi.edu/blog/hvacr/dedicated-outdoor-air-system/

Nimlyat, P. S., Salihu, B., & Wang, G. P. (2021). The impact of indoor environmental quality (IEQ) on
patients’ health and comfort in Nigeria. https://doi.org/10.1108/IJBPA-06-2021-0089

Nor, N. S. M., Yip, C. W., Ibrahim, N., Jaafar, M. H., Rashid, Z. Z., Mustafa, N., Hamid, H. H. A., Chandru,
K., Latif, M. T., Saw, P. E., Lin, C. Y., Alhasa, K. M., Hashim, J. H., & Nadzir, M. S. M. (2021). Particulate
matter (PM2.5) as a potential SARS-CoV-2 carrier. Scientific Reports 2021 11:1, 11(1), 1–6.
https://doi.org/10.1038/s41598-021-81935-9

NR Acoustic Comfort | NJ Green Building Manual. (n.d.). Retrieved August 31, 2022, from
https://greenmanual.rutgers.edu/acoustic-comfort/

Ojha, V. K., Griego, D., Kuliga, S., Bielik, M., Buš, P., Schaeben, C., Treyer, L., Standfest, M., Schneider,
S., Kö nig, R., Donath, D., & Schmitt, G. (2019). Machine learning approaches to understand the influence
of urban environments on human’s physiological response. Information Sciences, 474, 154–169.
https://doi.org/10.1016/J.INS.2018.09.061

Osibona, O., Solomon, B. D., & Fecht, D. (2021). Lighting in the Home and Health: A Systematic Review.

108
International Journal of Environmental Research and Public Health, 18(2), 1–20.
https://doi.org/10.3390/IJERPH18020609
Prabhakaran, S. (2022). Granger Causality Test in Python - Machine Learning Plus.
https://www.machinelearningplus.com/time-series/granger-causality-test-in-python/

Qian, H., & Zheng, X. (2018). Ventilation control for airborne transmission of human exhaled bio-aerosols
in buildings. Journal of Thoracic Disease, 10(19), S2295–S2304. https://doi.org/10.21037/JTD.2018.01.24

Ramaswamy, M., Al-Jahwari, F., & Al-Rajhi, S. M. M. (2010). IAQ in Hospitals - Better Health through
Indoor Air Quality Awareness. https://oaktrust.library.tamu.edu/handle/1969.1/94139

Random Forest Algorithm. (n.d.). Retrieved February 28, 2023, from
https://www.simplilearn.com/tutorials/machine-learning-tutorial/random-forest-algorithm

Renard, J. B., Surcin, J., Annesi-Maesano, I., Delaunay, G., Poincelet, E., & Dixsaut, G. (2022). Relation
between PM2.5 pollution and Covid-19 mortality in Western Europe for the 2020–2022 period. Science of
The Total Environment, 848, 157579. https://doi.org/10.1016/J.SCITOTENV.2022.157579

S E N S O R L I N E U P Sensors for Indoor and Outdoor Environmental Quality Humidity Air Pollution
Particle Formaldehyde Pollen. (n.d.).

Sadek, A., & Nofal, E. (2013). Effects of Indoor Environmental Quality on Occupant Satisfaction in
Healing Environments. Proceedings of Building Simulation Cairo 2013 Conference: Towards Sustainable
& Green Life. https://lirias.kuleuven.be/2802270

Sadrizadeh, S., Yao, R., Yuan, F., Awbi, H., Bahnfleth, W., Bi, Y., Cao, G., Croitoru, C., de Dear, R.,
Haghighat, F., Kumar, P., Malayeri, M., Nasiri, F., Ruud, M., Sadeghian, P., Wargocki, P., Xiong, J., Yu,
W., & Li, B. (2022). Indoor air quality and health in schools: A critical review for developing the roadmap
for the future school environment. Journal of Building Engineering, 57, 104908.
https://doi.org/10.1016/J.JOBE.2022.104908

Salami, A. (2022, February 19). Cross Correlation with Two Time Series in Python.
https://www.datainsightonline.com/post/cross-correlation-with-two-time-series-in-python

Song, B., Bai, L., & Yang, L. (2022). Analysis of the long-term effects of solar radiation on the indoor
thermal comfort in office buildings. Energy, 247, 123499.
https://doi.org/10.1016/J.ENERGY.2022.123499

Song, V. (2022). How to measure heart rate on your smartwatch - The Verge.
https://www.theverge.com/23004703/heart-rate-ekg-smartwatch

109
Sound Zero | How Acoustics Affect Productivity In Offices. (n.d.). Retrieved August 31, 2022, from
https://sound-zero.com/how-acoustics-affect-productivity-in-office-environments/
Sun, C., Huang, X., Zhang, J., Lu, R., Su, C., & Huang, C. (2022). The new model for evaluating indoor
air quality based on childhood allergic and respiratory diseases in Shanghai. Building and Environment,
207, 108410. https://doi.org/10.1016/J.BUILDENV.2021.108410

Tang, T., Zhang, Y., Zheng, Z., Zhou, X., Fang, Z., & Liu, W. (2021). Detailed thermal indicators analysis
based on outdoor thermal comfort indices in construction sites in South China. Building and Environment,
205, 108191. https://doi.org/10.1016/J.BUILDENV.2021.108191

Tatsat, H., Puri, S., Lookabaugh, B., & Safari, an O. M. Company. (2020). Machine Learning and Data
Science Blueprints for Finance. 400. https://www.oreilly.com/library/view/machine-learning-
and/9781492073048/

Thermal comfort in buildings - Designing Buildings. (n.d.). Retrieved August 31, 2022, from
https://www.designingbuildings.co.uk/wiki/Thermal_comfort_in_buildings

Tian, X., Cheng, Y., & Lin, Z. (2022). Modelling indoor environment indicators using artificial neural
network in the stratified environments. Building and Environment, 208, 108581.
https://doi.org/10.1016/J.BUILDENV.2021.108581

Torku, A., Chan, A. P. C., Yung, E. H. K., & Seo, J. (2022). Detecting stressful older adults-environment
interactions to improve neighbourhood mobility: A multimodal physiological sensing, machine learning,
and risk hotspot analysis-based approach. Building and Environment, 224, 109533.
https://doi.org/10.1016/J.BUILDENV.2022.109533

Understanding ESG: Environmental Factors | M&H Engineering & Consulting for The Oil & Gas
Industry. (n.d.). Retrieved January 31, 2023, from http://www.mhes.com/esg-environmental/

Waili, T., Sa’Ad Alshebly, Y., Azami Sidek, K., & Md Johar, M. G. (2020). Stress recognition using
Electroencephalogram (EEG) signal. Journal of Physics: Conference Series, 1502(1), 012052.
https://doi.org/10.1088/1742-6596/1502/1/012052

What is biometrics? (n.d.). Retrieved September 6, 2022, from
https://www.techtarget.com/searchsecurity/definition/biometrics

What Is Environmental, Social, and Governance (ESG) Investing? (n.d.). Retrieved January 31, 2023, from
https://www.investopedia.com/terms/e/environmental-social-and-governance-esg-criteria.asp

What is ESG (Environmental, Social & Governance)? (n.d.). Retrieved January 31, 2023, from
https://corporatefinanceinstitute.com/resources/esg/esg-environmental-social-governance/

110

What is Skin Temperature and Why Should You Monitor It? (n.d.). Retrieved September 8, 2022, from
https://www.whoop.com/thelocker/what-is-skin-temperature-and-why-should-you-monitor-it/

What is Supervised Learning? | IBM. (n.d.). Retrieved February 28, 2023, from
https://www.ibm.com/topics/supervised-learning

Wolkoff, P., Azuma, K., & Carrer, P. (2021). Health, work performance, and risk of infection in office-like
environments: The role of indoor temperature, air humidity, and ventilation. International Journal of
Hygiene and Environmental Health, 233, 113709. https://doi.org/10.1016/J.IJHEH.2021.113709

WONGTRAKOOL, B. M. W. (2018). ESG Essentials: What You Need to Know About Environmental,
Social and Governance Investing | Western Asset. Western Asset.
https://www.westernasset.com/au/qe/research/whitepapers/esg-essentials-what-you-need-to-know-2018-
04.cfm

Wu, Z., Li, N., Peng, J., & Li, J. (2019). Effect of long-term indoor thermal history on human physiological
and psychological responses: A pilot study in university dormitory buildings. Building and Environment,
166, 106425. https://doi.org/10.1016/J.BUILDENV.2019.106425

Wu, Z., Li, N., & Schiavon, S. (2022). Experimental evaluation of thermal comfort, SBS symptoms and
physiological responses in a radiant ceiling cooling environment under temperature step-changes. Building
and Environment, 224, 109512. https://doi.org/10.1016/J.BUILDENV.2022.109512

Yang, L., Yan, H., & Lam, J. C. (2014). Thermal comfort and building energy consumption implications –
A review. Applied Energy, 115, 164–173. https://doi.org/10.1016/J.APENERGY.2013.10.062

Yang, R., Yang, L., Wu, Z., & Jin, W. (2022). Physiological and perceptual responses due to exposure to
three indoor temperatures in cold environments. Energy and Buildings, 272, 112355.
https://doi.org/10.1016/J.ENBUILD.2022.112355

Abstract (if available)

Abstract During the COVID-19 pandemic, Indoor Environmental Quality (IEQ) was considered more seriously by the public because it pertained to the risk of infection. Especially for commercial buildings, the indoor environment was linked with occupants’ health and productivity as well as the company’s performance according to ESG criteria. However, even though there were many studies focusing on the evaluation of IEQ, the experiments were not comprehensive to develop a predictive model between human physiological responses and indoor environmental quality. This research investigated the correlations between human physiological responses and IEQ components in two office areas with 14 participants. The IEQ data, including indoor air quality, thermal comfort, lighting, and acoustics, were collected using sensors, while simultaneously wearable devices were used to record human physiological response data. Cross-correlation analysis was applied to establish the correlations between indicators when analyzing datasets from sensors and wearable devices. Then predicted models were generated for personal and general data by using supervised machine learning for IAQ and thermal evaluation based on human bio-signals. As a result, this research confirmed that the human physiological signals provided feasibility to predict the user’s environmental satisfaction as a function of the data-driven model at both the individual and the general models based on the data of multiple human subject experiments. Even though this study provided a robust conclusion based on the analyzed results of the dataset, the prediction performance could be better by accommodating more datasets with more participants from different ages and occupied environments for a future study.

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

Indoor air quality for human health in residential buildings

PDF

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

PDF

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

PDF

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

PDF

Multi-occupancy environmental control for smart connected communities

PDF

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

PDF

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

PDF

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

PDF

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

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

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

PDF

Impact of bamboo materials on indoor air quality

PDF

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

PDF

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

PDF

Developing a data-driven model of overall thermal sensation based on the use of human physiological information in a built environment

PDF

Using bio-signals with smart windows

PDF

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

PDF

Data visualization in VR/AR: static data analysis in buildings

PDF

Enhanced post occupancy evaluation (POE) for office building: improvement of current methodology to identify impact of ambient environment

PDF

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

Asset Metadata

Creator Dai, Haoyue (author)

Core Title Indoor environmental quality and comfort: IEQ adaptation and human physiological responses in commercial buildings

School School of Architecture

Degree Master of Building Science

Degree Program Building Science

Degree Conferral Date 2023-05

Publication Date 10/18/2024

Defense Date 03/29/2023

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

Tag bio-signals,human comfort,indoor environmental quality,OAI-PMH Harvest,thermal comfort

Format theses (aat)

Language English

Contributor Electronically uploaded by the author (provenance)

Advisor Choi, Joon-Ho (committee chair), Susman, Gideon (committee member), (Lerman, Kristina)

Creator Email daihy516@gmail.com,haoyueda@usc.edu

Unique identifier UC113038282

Identifier etd-DaiHaoyue-11648.pdf (filename)

Legacy Identifier etd-DaiHaoyue-11648

Document Type Thesis

Format theses (aat)

Rights Dai, Haoyue

Internet Media Type application/pdf

Type texts

Source 20230419-usctheses-batch-1025 (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