Dynamic workplace platform: exploration of the feasibility of human electroencephalogram (EEG) to predict the user’s indoor environmental satisfaction

PDF

Download

Open document

Flip pages

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content

DYNAMIC WORKPLACE PLATFORM:

Exploration of the Feasibility of Human Electroencephalogram (EEG) to Predict the User’s
Indoor Environmental Satisfaction

by

Xiaoyu Yin

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 am writing to express my sincere gratitude for all the support and assistance my parents
have provided me during my two-year study at USC. Without their invaluable help, I would not
have been able to complete my research project successfully.
I am also grateful to Professor Joon-Ho Choi for his guidance and mentorship, which helped
me to define and develop my research topic. His expertise in the field of architecture and
experience in the research were incredibly valuable in providing me with a solid foundation for
my work. His encouragement to be more creative and innovative during the research process was
truly inspiring.
I would like to express my appreciation to Professor Gideon Susman and Professor Kristina
Lerman for their valuable advice and expertise in different areas, and I am especially grateful to
Professor Gideon Susman for providing me with the site for this experiment and for his
enthusiasm in helping me solve problems during the experiment. I would also like to thank WSP
for supporting this experiment.
I also wish to express my gratitude to the volunteers who participated in this experiment.
Thanks to Saba Imani for helping me with R studio and Python during the subsequent
prediction modeling phase.

Joon-Ho Choi joonhoch@usc.edu
Gideon Susman gsusman@usc.edu
Kristina Lerman lerman@isi.edu

iii
Table of Contents
Acknowledgements ......................................................................................................................... ii
List of Tables ................................................................................................................................. vi
List of Figures ............................................................................................................................... vii
Abstract ........................................................................................................................................ viii
Chapter 1. Introduction ................................................................................................................ 1
1.1 Impacts of Indoor environmental quality (IEQ) on human factors ................................... 1
1.1.1 Acoustics .................................................................................................................... 2
1.1.2 Thermal comfort ........................................................................................................ 2
1.1.3 Visual comfort ........................................................................................................... 3
1.1.4 Indoor air quality (IAQ) ............................................................................................. 3
1.2 Human physiological responses commonly used in research ........................................... 3
1.2.1 Properties of brain signal ........................................................................................... 4
1.2.2 Skin temperature in relation to thermal comfort ........................................................ 5
1.2.3 Changes in heart rate and heart rate variability during thermal adaptation ............... 6
1.3 The birth of federated learning ......................................................................................... 7
1.3.1 The basic framework of federated learning ............................................................... 8
1.3.2 Security for Federated Learning ................................................................................ 9
1.4 Dynamic workplace platform ......................................................................................... 10
1.5 Summary ......................................................................................................................... 12
Chapter 2. Literature review ...................................................................................................... 14
2.1 The relationship between IEQ and brain signals ............................................................ 14
2.1.1 Effect of temperature ............................................................................................... 14
2.1.2 Effect of color .......................................................................................................... 15
2.1.3 Overall consideration ............................................................................................... 15
2.2 Federal learning in the architectural industry ................................................................. 16
2.2.1 Building energy consumption projections ............................................................... 17
2.2.2 Potential application federated learning for smart city development ...................... 18
2.2.3 Summary of the role of federated learning .............................................................. 20
iv
2.3 The influencing factors of environmental comfort ......................................................... 20
2.3.1 Thermal, luminous, and acoustic environments have an impact on indoor comfort 20
2.3.2 Effects of subjective and non-physical parameters .................................................. 22
2.4 Potential use of EEG in the building environmental study ............................................. 24
2.4.1 Feedback collection method combining virtual reality and EEG ............................ 24
2.4.2 EEG-based to enhance human-build interaction ...................................................... 25
2.4.3 Validating the benefits of sustainable building with EEG and virtual scene ........... 26
2.4.4 Potential problems with EEG studies ...................................................................... 27
2.5 Summary ......................................................................................................................... 27
Chapter 3. Methodology ............................................................................................................ 29
3.1 Experimental equipment ................................................................................................. 31
3.1.1 OMEGA HHLM-1 Digital Lightmeter .................................................................... 32
3.1.2 HOBO MX1102 carbon dioxide logger ................................................................... 32
3.1.3 PCE-SDL 1 .............................................................................................................. 33
3.1.4 Emotiv Insight 2.0 .................................................................................................... 34
3.1.5 Empatica Embrace 2 ................................................................................................ 35
3.1.6 Garmin Vivosmart 3 ................................................................................................ 36
3.2 Experiment procedure ..................................................................................................... 37
3.3 Data processing ............................................................................................................... 39
3.4 Summary ......................................................................................................................... 41
Chapter 4. Data analysis and results .......................................................................................... 42
4.1.1 Participant 1 ............................................................................................................. 45
4.2 ANOVA analysis ............................................................................................................ 54
4.2.1 Overall IEQ satisfaction and EEG data ................................................................... 56
4.2.2 Thermal satisfaction and EEG data .......................................................................... 58
4.2.3 Background acoustic satisfaction and EEG data ..................................................... 60
4.2.4 Illumination intensity satisfaction and EEG data ..................................................... 61
4.2.5 Indoor air quality satisfaction and EEG data ........................................................... 63
4.2.6 Spatial quality satisfaction and EEG data ................................................................ 64
v
4.2.7 Work efficiency satisfaction and EEG data ............................................................. 66
4.3 Results ............................................................................................................................. 67
4.3.1 Summary of time lag cross correlation analysis ...................................................... 67
4.3.2 Summary of ANOVA analysis ................................................................................ 70
4.4 Summary ......................................................................................................................... 70
Chapter 5. Prediction model ...................................................................................................... 71
5.1 Establishment of support vector machine model ............................................................ 71
5.1.1 Predictive model algorithm selection ....................................................................... 72
5.1.2 Algorithm Explanation ............................................................................................. 73
5.2 Application of decision tree model ................................................................................. 76
5.2.1 Decision Tree algorithm explanation ....................................................................... 77
5.2.2 Sample decision tree ................................................................................................ 78
5.3 Reliability test of the prediction model ........................................................................... 80
5.4 Summary ......................................................................................................................... 82
Chapter 6. Conclusion and future work ..................................................................................... 83
6.1 Discussion ....................................................................................................................... 83
6.1.1 Evaluation of the workflow ..................................................................................... 85
6.2 Research Limitations ...................................................................................................... 86
6.2.1 Limitations of the tools ............................................................................................ 86
6.2.2 Methodological limitations ...................................................................................... 87
6.3 Future work ..................................................................................................................... 88
6.3.1 Potential research directions .................................................................................... 89
6.4 Summary ......................................................................................................................... 91
References: .................................................................................................................................... 92

vi
List of Tables
Table 2.1The data type of the monitoring sensor ................................................................... 23
Table 3.1 Sensor Information ................................................................................................. 36
Table 3.2 Occupant Satisfaction Questionnaire ..................................................................... 38
Table 3.3 Data sample from the first volunteer ...................................................................... 39
Table 4.1 Cross-correlation analysis without time lag ........................................................... 47
Table 4.2 Cross-correlation analysis with 1-minute lag ......................................................... 48
Table 4.3 Cross-correlation analysis with 2-minute lag ......................................................... 50
Table 4.4 Cross-correlation analysis with 3-minute lag ......................................................... 51
Table 4.5 Cross-correlation analysis with 4-minute lag ......................................................... 52
Table 4.6 Cross-correlation analysis with 5-minute lag ......................................................... 53
Table 4.7 Integration of time lag cross correlation analysis results ....................................... 54
Table 4.8 Correlation analysis between EEG and environmental data .................................. 68
Table 4.9 Number of people applicable at different time intervals ........................................ 69
Table 4.10 ANOV A analysis summary .................................................................................. 70

vii
List of Figures
Figure 1.1 Types of brain waves .............................................................................................. 4
Figure 1.2 Federal Learning Workflow .................................................................................... 9
Figure 1.3 Dynamics workplace platform ............................................................................. 11
Figure 2.1 Federal learning based building energy prediction architecture ........................... 18
Figure 2.2 The room layout ................................................................................................... 21
Figure 2.3 Satisfaction survey form ....................................................................................... 21
Figure 2.4 The relevance of the built environment to human inhabitants ............................. 25
Figure 3.1 Workflow overview .............................................................................................. 30
Figure 3.2 TK1SC floor plan ................................................................................................. 30
Figure 3.3 OMEGA HHLM-1 Digital Lightmeter ................................................................. 32
Figure 3.4 HOBO MX1102 carbon dioxide logger ............................................................... 33
Figure 3.5 PCE-SDL 1 ........................................................................................................... 34
Figure 3.6 Emotiv Insight 2.0 ................................................................................................ 35
Figure 3.7 Empatica Embrace 2 ............................................................................................. 36
Figure 3.8 Vivosmart 3 .......................................................................................................... 37
Figure 3.9 Timeline ................................................................................................................ 37
Figure 4.1 Time lag cross correlation visualization model .................................................... 44
Figure 4.2 Time lag cross correlation analysis setup diagram ............................................... 46
Figure 4.3 Minitab data analysis settings ............................................................................... 56
Figure 4.4 ANOV A analysis of overall indoor environmental satisfaction and EEG data .... 57
Figure 4.5 ANOV A analysis of thermal satisfaction and EEG data ....................................... 59
Figure 4.6 ANOV A analysis of background acoustic satisfaction and EEG data .................. 60
Figure 4.7 ANOV A analysis of illumination intensity satisfaction and EEG data ................. 62
Figure 4.8 ANOV A analysis of IAQ and EEG data ............................................................... 63
Figure 4.9 ANOV A analysis of spatial quality satisfaction and EEG data ............................ 65
Figure 4.10 ANOV A analysis of work efficiency satisfaction and EEG data ........................ 66
Figure 5.1 The Coding Script of the SVM algorithm ............................................................ 74
Figure 5.2 Data Selection ....................................................................................................... 75
Figure 5.3 Data splitting ........................................................................................................ 75
Figure 5.4 SVM model training ............................................................................................. 75
Figure 5.5 Accuracy prediction .............................................................................................. 76
Figure 5.6 Library import ...................................................................................................... 77
Figure 5.7 Decision tree variables creation ............................................................................ 78
Figure 5.8 Decision tree classifier generation ........................................................................ 78
Figure 5.9 Accuracy value export .......................................................................................... 78
Figure 5.10 EEG and thermal satisfaction decision tree ........................................................ 79
Figure 5.11 Scatter Plot of Predicted vs. Actual Thermal Satisfaction Scores ...................... 81

viii
Abstract:
As society becomes increasingly aware of the importance of indoor environmental quality (IEQ),
the need for more comprehensive assessments of IEQ is growing. With the advent of new types
of offices, such as dynamic office platforms, there is a pressing need to assess the impact of IEQ
on human physiology. More studies need to be used to assess indoor air quality (IEQ). This study
aims to investigate whether differences in human physiological responses are caused by
variations in IEQ in different areas of the same office. Data collection was conducted at an
innovative engineering firm in Southern California that already started adopting a dynamic work
platform. This research collected brain signals from employees working in different areas of the
office and analyzed the relationship between their EEG signals and various indoor environmental
parameters, as well as their responses to questionnaires about their perceptions of the indoor
environment. The results showed the influence of EEG signals and the indoor environment that
can be visualized. The results of the analysis could be used to develop a prediction model that
can be used to visualize the influence of EEG signals on indoor environmental quality. This
model will be applied to predict future indoor environmental parameters based on existing EEG
data, allowing people to effectively improve the indoor environment in the future. The study
found that EEG signals can be used to determine indoor temperature, but further research is
needed to investigate the relationship between other physiological responses and indoor
environmental parameters. This will lead to the development of a more comprehensive model for
predicting and improving indoor environmental quality.

Keywords: electroencephalography, indoor environmental quality, prediction model

Hypothesis: By examining the relationship between EEG signals and the indoor environment
quality parameters such as temperature, CO2 level, and acoustic, a predictive model, which
accurately predicts an individual's satisfaction level with their indoor environment, can be
developed to provide a more comfortable working environment for building occupants through
the building management system.

Research Objectives:
 To establish a direct link between EEG signals and real-time indoor environmental
parameters
 To analyze the correlation between the EEG signal and each question in the employee
satisfaction survey
 To build predictive models using existing data
 To test the accuracy of the predictive model

1
Chapter 1. Introduction
Indoor environmental quality has become an effective evaluation method. However, if
changes are made to the IEQ after the building has been put into use, it will not only waste time
and money but also cause environmental pollution. Therefore, if virtual reality technology is
used to simulate the environment and get user feedback, this problem can be effectively solved
prior to construction. Simultaneously, suppose the physiological responses of occupants can be
collected and used to establish a relationship with IEQ. In that case, designers will be able to
evaluate indoor environmental quality from subjective and objective aspects in the future.

1.1 Impacts of Indoor environmental quality (IEQ) on human factors
In recent decades, research has continued on how indoor environmental conditions affect
office workers' performance, health, and satisfaction. Providing an environment that meets most
occupants' comfort requirements is essential. It has been a primary goal of traditional facility
management practices, particularly in commercial office environments where individual control
over their surroundings is often limited (Kim et al., 2013).
Over the years, experts in different fields, such as physiologists, architects, engineers,
occupational health, and industrial hygiene, have been working on the four components of indoor
environmental quality: acoustics, thermal comfort, visual comfort, and indoor air quality
(D’Ambrosio Alfano et al., 2014). While these elements may seem simple and obvious, they
produce a distinctive response when they work together in a building. For example, a high-
quality indoor environment can make users happy and improve work efficiency; on the contrary,
2
it will make employees uncomfortable and even cause physical harm. Therefore, it is crucial to
understand the complex effect of IEQ on individuals in indoor spaces.

1.1.1 Acoustics
The most common issues associated with acoustics in the office are privacy and disruption
(Cowan, 2014). The reasons are that distractions can affect productivity, and privacy concerns
have recently gained wider attention due to numerous privacy protection laws passed by the US
federal government and the European Parliament. Using sound-absorbing materials for office
walls and roofs is helpful because it can prevent sound from being transmitted over considerable
distances and causing remote interference. At the same time, the partition can be appropriately
used to block the view of others to achieve the purpose of privacy.

1.1.2 Thermal comfort
Thermal comfort is the degree of satisfaction with the thermal environment, which is a
subjective evaluation; however, this situation can also be assessed by objective adjustment,
which is to think of the human body as a thermodynamic system exchanging heat with the
physical environment around it (D’Ambrosio Alfano et al., 2014). At the same time, thermal
comfort also needs to meet another condition, local thermal discomfort (Parsons, 2007). In other
words, there are temperature differences in certain room areas due to vertical air temperature
differences.

3
1.1.3 Visual comfort
The level of illumination significantly affects the physiology and psychology of the
occupants. People cannot perform their daily activities effectively, efficiently, and comfortably
without appropriate lighting. In addition, the lack of regular light exposure can cause a range of
adverse symptoms for employees, such as fatigue and stress.

1.1.4 Indoor air quality (IAQ)
Poor indoor air quality in buildings can reduce productivity, in addition to causing
employees to express dissatisfaction. For example, poor indoor air affects most aspects of office
performance by as much as 6-9% (Wyon, 2004).
The space, which humans occupy, has a critical impact on our physical and mental health,
affecting the quality of decisions we make, how we feel internally at work, how we move around
the work area and interact with others, and the ability to perform daily tasks. IEQ has become a
significant area of concern for the building, engineering, and construction industries because it is
critical to the U.S. Green Building Council's LEED Green Building rating system. Many
construction projects have targeted green rating systems because they add value to the project,
demonstrate a commitment to sustainability and ethical building, and keep design at the cutting
edge (Anonymous, 2022).

1.2 Human physiological responses commonly used in research
The physiological response is generally when the outside world stimulates an individual
4
after a response. There are many examples in life, such as a person's stomach growling when he
feels hungry and closing his eyes involuntarily when sand flies into his eyes. Heart rate, blood
glucose, blood pressure, respiratory rate, body temperature, blood volume, sound pressure,
photoplethysmography, electroencephalogram, electrocardiogram, oxygen saturation, and skin
conductance are common physiological data that wearable devices can collect (Perez-Pozuelo et
al., 2021).

1.2.1 Properties of brain signal
The brain signal, also known as an electroencephalogram (EEG), is the physiological
parameter of this experiment. EEG is the current medical recording method for brain function
research. Figure 1.1 records the four simple periodic rhythms: alpha(α), beta(β), delta(δ), and
theta(θ) (Yao et al., 2008). Each brainwave has its corresponding different state of brain
consciousness. It can also be said that different brainwaves are needed in different states of
consciousness to best accomplish the work of the brain. Just like the gears of a car, each gear has
its appropriate state.

Figure 1.1 Types of brain waves (Yao et al., 2008)

5
The properties of EEG signals can be described as complex. The complexity of EEG stems
from the intricacies of the nervous system. Spontaneous EEG is described as a linear stochastic
process that closely resembles noise. From a signal processing perspective, EEG has noise and
pseudo-randomness, time-varying and non-smoothness, and high non-linearity (Thakor & Tong,
2004). First, EEG is usually between 10 and 300 μV and is susceptible to a variety of
physiological and electrical noises. Even the EEG shows a high degree of randomness and non-
smoothness. It is why the second characteristic of brain waves mentions that it is not a smooth
process but varies with the physiological state. The waveforms of EEG can include regular sine
waves, irregular spikes/multi-spikes, or fusiform/multi-fusiform complex waves. For example,
during seizures in epileptic patients, their EEG may show significant oddness or non-
smoothness. However, in practice, the EEG is a relatively short resting process. Finally, although
EEG is a nonlinear process, it is nonetheless time-, state-, and location-dependent.
In addition, inferences of emotional states from EGG have received considerable attention
because EEG can directly reflect emotional states and is easy to measure (M.-K. Kim et al.,
2013).

1.2.2 Skin temperature in relation to thermal comfort
When humans are exposed to changes in the state of the thermal environment, they adapt to
the environment through metabolism and thermoregulation to compensate for the heat transfer
generated between the core of the body and the surrounding environment (Noda et al., 2018).
Human thermoregulation is a process that occurs when the body produces excess body heat
6
resulting in an increase in the body's core temperature. It is when the thermoregulatory center in
the hypothalamus transmits body heat through the blood and tissues to the skin. When the heat
reaches the skin's surface, it is dissipated by vasodilation and perspiration. Eventually, the body
will gradually return to homeostasis. However, if the temperature sensed by the temperature
receptors suddenly changes significantly and is accompanied by a large heat transfer, then the
thermoregulatory system does not respond effectively to thermoregulation, and to lead to
discomfort (C.-P. Chen et al., 2011). For example, in summer, when people leave the cool air-
conditioned room and come to the hot outdoor area, the sudden temperature will make the body's
thermoregulation function fail, thus causing discomfort.
The human skin temperature should be steadily changing in a stable thermal environment.
Therefore, it is helpful to use skin temperature as one of the physiological indicators for
assessing thermal comfort.

1.2.3 Changes in heart rate and heart rate variability during thermal adaptation
In integrative physiology, heat acclimation features include increased sweating and
diminished heart rate (Flouris, 2011). The analysis of HRV is often used to study and assess the
status of body systems. HRV is primarily influenced by the degree of tension in the body's
regulatory system. Tension in the regulatory system is the body's response to the entire complex
system of indoor environments that affect it.
The stability of the human heart rate will be affected, especially in an office with insufficient
indoor environmental quality. It is because a healthy person has sufficient capacity to regulate
7
work tension in response to stressors through the regulatory system. Therefore, in a good
working environment, the human heart rate will show a stable state. On the contrary, if the
regulatory system is unable to make timely adjustments in a poor environment, the stability of
the human heart rate will be broken.

1.3 The birth of federated learning
Nowadays, people have entered the era of big data, and almost all the information in daily
life is transmitted through the Internet. Therefore, people are gradually realizing the importance
of data security. Although Internet companies have taken many measures for user information
security, there is still the problem of privacy and confidential information leakage. With the
development of computer technology, machine semester has become the most common
technique used for data analysis and processing in this period. However, it faces two challenges.
On the one hand, ensuring the security of the data used in machine learning is difficult,
which is unacceptable in today's privacy-conscious society. On the other hand, the emergence of
the Internet of Things has facilitated data sharing, and if data protection is tightened among
enterprises to prevent data leakage, the amount of data that can be shared will be reduced. The
solution to these challenges is Federated Learning, a novel distributed learning framework that
allows multiple participants to share training results without compromising the privacy of their
data (Yang et al., 2022).

8
1.3.1 The basic framework of federated learning
The increasing use of IoT platforms and smart devices has led to an increase in the amount
of data and accelerated the adoption of machine learning into more domains. However, privacy
concerns have also grown in parallel, leading to a shortage of data for machine learning systems.
At the same time, many countries have introduced a series of data protection measures in recent
years, and data privacy has become one of the most important ethical principles for machine
learning systems (Jobin et al., 2019). Moreover, raw data often needs pre-processing before it
can be used for model training.
In order to address these issues, a decentralized approach called federated learning has
emerged. This approach distributes the training data across multiple mobile devices, allowing
shared models to be learned by aggregating updates from local computations (McMahan et al.,
2017). In this way, data privacy is preserved while allowing more data to be used in machine
learning applications.
The conceptual framework of federated learning, as depicted in Figure 1.2, consists of three
main components: learning coordinators, contributor clients, and customers (Lo et al., 2022). The
learning coordinator serves as the owner of the federated learning system, while the contributor
client represents the individual or organization that provides the system data and trains the local
model. The customer is the user of the software or smart device.
9

Figure 1.2 Federal Learning Workflow

At the same time, according to the diagram, the training data in the federated learning
process remains locally stored on the client's device. The workflow of the federated learning
process involves the client downloading the global model from the server. Subsequently, when
the client generates new data using the app or smart device, the client will commence training on
this local data. The results of the data are eventually uploaded from the user's client to the server
and eventually aggregated with other users’ data in the central server to update the model.

1.3.2 Security for Federated Learning
As healthcare costs continue to rise, populations age forest talk, universal health coverage,
and new epidemics like COVID-19 spread across the world, the sharing of healthcare data
becomes increasingly critical. However, medical data, by design to patient privacy issues, has led
to data silos and insufficient data volume. This problem cannot be solved by data from just one
hospital. Currently, medical data faces several challenges, such as the limited amount of data that
can be used to train models, the variability in data availability among hospitals and institutions,
10
and the differences in disease types, ethnicities, and gender data held by each hospital (H. Li et
al., 2023).
While machine learning can help improve the efficiency of clinical trials and decision-
making processes, traditional machine learning training models require users to send personal
information directly to servers, which increases the risk of privacy breaches. However, federated
learning offers a solution by allowing data owners to jointly train neural networks without
sharing personal data (J. Chen et al., 2023). The result is that it can share sensitive data generated
from IoT devices and perfectly address the patient privacy concerns of the healthcare industry.

1.4 Dynamic workplace platform
From a historical perspective, office configurations have evolved in conjunction with
advances and developments in building technology, such as the open-plan offices (OPOs) that
have been the most common type of office in the United States since the early 20th century
(Babapour et al., 2018). However, the drawbacks of this type of office, such as space congestion
and privacy and security issues, have gradually emerged over time. Especially during COVID-
19, this led to severe consequences due to the proximity of people in this office style. In this
particular case, the new work platform will play a crucial role in order to be able to protect the
safety of the workers better. The rapid development of information technology makes dynamic
work platforms possible (Wohlers & Hertel, 2017).
Through analysis of the company in this research, it was found that a dynamic workplace
platform, illustrated in Figure 1.3, is a space that can be easily reconfigured to meet the
11
company's needs through furniture, collaborative spaces, and technology. Such a design is
intended to better meet the needs of employees.

Figure 1.3 Dynamics workplace platform

One of the critical reasons dynamic working platforms are loved is that they increase the
flexibility of the office area (Rolfö, 2018). Workplace flexibility means that companies can give
their employees flexibility, which is crucial in today's fast-paced workplace because it reduces
unnecessary movement at work. Employees can switch spaces without impeding colleagues'
personal space or taking up specific workspaces. In other words, a dynamic workplace platform
allows the physical workplace to keep up with the needs of employees in real time, whether they
need a specific type of workstation or need to relocate multiple employees on short notice. It also
gives them the power to choose seats and increase their comfort level. Allowing employees to
choose their seats or workstation may seem simple. However, from another point of view, the
company believes employees perform better in the most comfortable working environment. At
the same time, the ability to choose where and how they work allows employees to choose the
area where they feel most comfortable. They choose the most efficient way when they are not
12
forced to adapt to a particular way of working. It is an appropriate method for both employees
and employers. It is a simple way to increase productivity.
Another advantage is that it promotes employee cooperation (Rolfö, 2018). Restrictions on
work areas are removed from this concept that the social work environment should be freely
distributed. Whether employees enjoy camaraderie or need to collaborate on a project, they will
have more freedom to work in sync when nothing ties them to a specific table or seating area. At
the same time, flexible seating allows teams to adapt to the needs of their work in real time. For
example, when a task requires employees from different departments to work together, they can
choose a separate area to work in completely. In the work process, if they need to communicate
with other team members, they can communicate with their team members immediately.
Finally, dynamic workplace platforms can bring about a change in the landscape. Monotony
is a productivity killer. Letting employees choose their desk arrangement for the day allows them
to break the tedium of the same seating assignments daily. A new perspective may trigger new
emotions and promote motivation, improving morale and productivity.

1.5 Summary
In order to enhance IEQ, it is a practical approach to obtain the physiological responses of
employees working in a building. While attention has been paid to IEQ in many office
environments, there are still areas that require improvement despite meeting the needs of most
employees. It is helpful to find areas that need improvement in IEQ by collecting brain signals
from employees in different work areas. Furthermore, establishing a direct relationship between
13
EEG signals and indoor environmental parameters is the foundation of this research. Only by
understanding this relationship can more complex prediction models be built upon it. Finally,
federated learning was selected for the prediction model over traditional machine learning due to
the personal privacy of the volunteers involved in collecting EEG signals. Thus, federated
learning is a better choice.

14
Chapter 2. Literature review
2.1 The relationship between IEQ and brain signals
It is well known that IEQ can significantly impact human health and well-being,
productivity, and cognitive performance. For example, poor air quality and high carbon dioxide
levels have been shown to impair cognitive function and increase fatigue. In addition, noise
pollution can interfere with cognitive performance and negatively affect mental health.
On the other hand, optimal IEQ conditions, such as good air quality and appropriate lighting,
can positively impact brain function and cognitive performance. For example, natural light and
good ventilation have improved productivity and reduced stress levels. Therefore, understanding
the relationship between IEQ and brain signals is critical to creating a healthy indoor
environment that promotes cognitive function, productivity, and overall health.

2.1.1 Effect of temperature
When humans are in different environments, their comfort levels can be different. When
people are in a relatively comfortable environment, they will perform better than in other
environments. For example, when college students study at a room temperature of 25.7℃, they
have the highest learning state, but their academic performance will decline under high or low
indoor temperatures (Wang et al., 2019). However, when the mental load index calculated from
the frontal θ and parietal α band power is used to test the mental load index of employees under
different thermal environments, the results will be different. The results suggest that the effects
15
of heat vary from person to person, with slightly warmer temperatures leading to better mental
functioning than other temperatures (H. Kim et al., 2020).

2.1.2 Effect of color
The color of the building environment also affects the brain signals of employees. The most
common colors are cool and warm colors in art. Warm colors such as red and yellow give people
a feeling of liveliness, cheerfulness, and excitement. However, cool colors like blue and green
can make people feel quiet and composed. For example, when people are placed in black, white,
and colored rooms, their emotional processing and brain activity also differ. When exposed to the
colored room, there is not only an increase in respiration but also a significant increase in alpha
frontal midline power and frontal hemisphere lateralization relative to the other conditions and
an increase in the power spectral density of theta, alpha, and beta bandwidths for all electrodes in
the blue condition (Bower et al., 2022). Thus, using colors to decorate office areas can make
employees feel depressed, which is inappropriate for brain activity. If various colors can be used
to decorate the building environment properly, employees' working conditions will be improved
accordingly.

2.1.3 Overall consideration
In addition to the temperature and color of a building affecting the brain, air quality and
sound also affect brain activity. Stress levels were highest when occupants were exposed to
temperatures of 30 °C, odor irritants (VOC), and road traffic noise, and differences in brain
16
mapping of relatively high beta waves in the temporal lobe were found to help determine the
stress status of participants (Choi et al., 2015). An overall poor indoor environment will only
increase the stress level of the occupants. It is vital to provide a comfortable environment for the
occupants to relax to keep them in good condition.
Human comfort will affect health and work performance. It largely depends on the indoor
environment. Temperature, hearing, vision, and air quality all affect the indoor environment and
can also significantly affect this attained activity. The total EEG energy in the parietal lobe is
smaller in a comfortable environment than in an uncomfortable one. When subjects were
comfortable, the theta band in the frontal pole increased significantly.
IEQ parameters are interdependent and must be considered together. Currently, many ways
to determine IEQ are subjective, such as questionnaires. Brain signals, however, provide the
most objective data. Through the brain signal data, we can get the most intuitive answer without
doing any questionnaire. This method not only saves time but also is convenient.
However, although many experiments have demonstrated a connection between IEQ and
EEG, these experiments have only studied one or several factors of indoor environmental quality.
If more parameters can be included, the experimental results will be more comprehensive.

2.2 Federal learning in the architectural industry
Many cities are investing in smart city infrastructure development to ensure sustainable
urban development. Building energy consumption is an important part of smart city
17
infrastructure development. If designers can predict building energy consumption, the result
plays a crucial role in achieving energy efficiency and sustainability goals for urban areas.
Traditional machine learning models require large amounts of data to accurately train the
model. However, privacy concerns and limited data availability make the traditional approach
challenging for building energy consumption prediction. In order to address these issues, more
urban planners are turning to federated learning, a decentralized machine learning approach.
Federated learning allows for the training of models without sharing sensitive data, making it an
ideal solution for building energy consumption prediction in the context of smart city
development.

2.2.1 Building energy consumption projections
Building energy management systems (BEMS) play an essential role in improving building
energy efficiency and reducing energy consumption, ultimately helping to achieve net-zero
energy buildings with low carbon emissions. Building load forecasting is one of the most critical
components of building control and analysis activities and grid interactions, and it forms the
foundation of building energy management systems (Zhang et al., 2021).
With high levels of intermittent generation and dynamic demand patterns, accurate
forecasting of residential loads becomes critical (Fernández et al., 2022). Smart meters provide
detailed load data and play a vital role in making these forecasts. Figure 2.1 depicts the general
architecture of a Federated Knowledge Sharing Framework, which is based on the data provided
by smart meters and uses the Federated Learning Methodology (Tang et al., 2023).
18

Figure 2.1 Federal learning based building energy prediction architecture

In the model, the building energy management system processes the data collected by the
smart meters to obtain data that can be directly and securely computed locally. However, this
data is aggregated to a central processor for machine learning. Ultimately, federal learning builds
a personalized predictive model for each building. This approach addresses privacy concerns and
ensures that data is secure and not shared with a central server. The model uses local data to build
a personalized predictive model that can be used to forecast the building's energy consumption
accurately.
In summary, the Federated Knowledge Sharing Framework enables building energy
management systems to achieve high levels of accuracy and efficiency while maintaining data
privacy and security.

2.2.2 Potential application federated learning for smart city development
In recent decades, the process of urbanization has resulted in a plethora of problems,
including urban poverty, high costs, traffic congestion, housing shortages, financial constraints,
19
rising crime rates, environmental degradation, and inequality (Ramu et al., 2022). Meanwhile,
smart cities rely heavily on sensors and Internet of Things devices to collect and track
information about vehicle traffic, waste, water, drainage, potholes, smart buildings, smart grids,
theft, and environmental monitoring(Jiang, 2020). These problems can be solved to some extent
by machine learning to analyze past data.
In addition, some smart city applications, such as traffic prediction, smart industry,
environmental monitoring, and disaster management, require real-time decision making, which
can be hindered by latency issues in cloud-based storage, particularly given the large amounts of
data generated by IoT devices (Pandya et al., 2023).
Federated learning is a novel machine learning approach that addresses some of these issues
by offloading global machine learning algorithms to devices rather than transferring raw data to a
central model. At the same time, federated learning helps to overcome privacy protection and
significant data processing issues, facilitating real-time decision making in smart cities.
Finally, federated learning can help smart cities to develop more accurate and personalized
energy-saving recommendations for building occupants. By analyzing the patterns of energy
consumption in different buildings, the system can identify areas for improvement and suggest
ways to reduce energy usages, such as turning off lights when a room is unoccupied or adjusting
the temperature settings to reduce unnecessary heating or cooling.

20
2.2.3 Summary of the role of federated learning
In summary, federated learning has the potential to revolutionize the way we predict
building energy consumption and develop smart cities. By overcoming the challenges of
traditional machine learning models, federated learning offers a way to predict building energy
consumption while preserving data privacy and decentralizing data storage.
However, to enhance occupants' experience, further analysis of user habits is required. By
recording physiological data such as heart rate, federated learning can be used to analyze how to
configure indoor environmental parameters to create a more comfortable environment for users.
This approach will provide a more personalized experience for users and further contribute to the
development of smart cities.

2.3 The influencing factors of environmental comfort
As society continues to develop and people's access to resources increases, there is a
growing demand for higher standards of living, including improved indoor environmental quality
(Kahneman & Deaton, 2010). It means people have a higher requirement for indoor environment
comfort. The comfort of living has also become an essential reference factor for people to
evaluate indoor environment quality.

2.3.1 Thermal, luminous, and acoustic environments have an impact on indoor comfort
Environmental factors comprehensively impact the acceptability and work efficiency of the
occupants. Tsinghua University in Beijing conducted a controlled field survey to explore their
21
interactions. The temperature, lighting, and noise level can be controlled in the experimental
place, as shown in Figure 2.2. The subjects were all college students. Half of them were men,
and half were women.

Figure 2.2 The room layout (Huang et al., 2012)

Different kinds of instruments were used in the experiment to measure environmental
variables. At the same time, the volunteers' personal information, including gender and clothing
conditions, was determined by a questionnaire. After completing the preparation, the volunteers
were asked to investigate their overall satisfaction with the indoor thermal environment, light
environment, sound environment, and indoor environment in the way shown in Figure 2.3.

Figure 2.3 Satisfaction survey form (Huang et al., 2012)

The results show that the three factors in the experiment have different effects on the overall
indoor environment, but the satisfaction of temperature and noise has veto power on the overall
satisfaction of the indoor environment (Huang et al., 2012).

22
2.3.2 Effects of subjective and non-physical parameters
In addition to the actual indoor environment conditions that can be monitored for physical
parameters, the perception and attitude of the occupants towards these conditions will also affect
their perception of comfort. In order to more comprehensively assess the indoor comfort
conditions in different areas of a factory, the experiment measured the indoor comfort conditions
in all four seasons of the factory by combining monitoring the physical environmental
parameters of the factory and submitting questionnaires to factory employees.
A particular indoor microclimate station was used in the experiment to monitor indoor
physical data, as shown in Table 2. At the same time, the questionnaire provided to the workers
in the experiment helped them understand their views and tolerance of different environmental
parameters and investigate how personal attitudes, behavior habits, and feelings influenced their
comfort level.
23

Table 2.1The data type of the monitoring sensor (Castaldo et al., 2018)

According to the study, employees are pleased with their work environment in terms of
physical parameters, and 79 percent believe that the company's design of a pleasant work
environment positively affects their judgment of comfort (Castaldo et al., 2018). Therefore, in
the working environment, companies can not only change the psychological factors of
employees by changing corporate policies but also control the quality of the indoor environment
and improve the overall comfort of employees.
To measure residents' satisfaction with environmental comfort, many experiments have
utilized questionnaires as a survey method. However, this approach is a process that results from
24
the subjective evaluation of the occupants. This process may be influenced by other factors, such
as real-name investigations. On the contrary, the brain signals can objectively reflect the
employee's satisfaction in the indoor environment.

2.4 Potential use of EEG in the building environmental study
When the indoor environment changes, it is difficult for humans to accurately describe their
subjective emotions, such as their level of preference and sense of stress. In fact, human
physiology and psychology both influence their evaluation of the built environment. Some
scholars have already started to evaluate the relationship between human feelings and the built
environment through brainwave data.

2.4.1 Feedback collection method combining virtual reality and EEG
Li et al. explored the correlation between spatial environment and human physiological,
psychological, and productivity by creating immersive spatial experiences using VR technology,
EEG signal collection, and laboratory environmental control techniques (Figure 2.4) (J. Li et al.,
2020).
25

Figure 2.4 The relevance of the built environment to human inhabitants (J. Li et al., 2020)

During the experiment, volunteers will experience different immersive virtual spaces after
using VR and EEG devices. During the experiment, their brain information data will be recorded.
At the end of the experiment, volunteers will also be given a subjective questionnaire to assess
their psychological perceptions. The experiment will end with analyzing human brain data to
construct the relationship between the effects of the spatial environment on human physiology
and work efficiency.

2.4.2 EEG-based to enhance human-build interaction
Human-architecture interaction has become an emerging discipline that is widely studied.
Engineers are incorporating digital information technology into building structures to face the
dynamic challenges of future living (Dalton et al., 2016). Human-architecture interaction is not
only the effect of the built environment on the occupants but also the feedback from the
occupants to the building.
26
Although the effects of indoor air quality on occupancy have been extensively studied, there
are limited ways for occupants to provide feedback to the building, such as direct adjustments by
occupants to the environmental quality in their area or improvements based on subjective
questionnaires (Melikov, 2004). Shan et al. control the indoor environment based on EEG
information and other real-time environmental data sensors to achieve human-building
information interaction (Shan et al., 2018). The experiment was conducted by comparing data
collected from questionnaires and brain signals regarding indoor environmental quality and self-
perception. The relationship between the two is established through machine learning, enhancing
interaction between humans and buildings.

2.4.3 Validating the benefits of sustainable building with EEG and virtual scene
Case studies have consistently shown the potential of sustainable buildings to improve
occupant well-being and productivity, but these conclusions have been reached through occupant
self-assessment. However, there is plenty of evidence that when people are out in nature, it can
bring health benefits, including reduced stress and improved mood (Kaplan, 1995). Today's
sustainable building designs also strive to create beautiful natural landscapes, appropriate indoor
air quality, and protect the environment.
In Hu et al.'s experiments, they used virtual environments, EEG, and event-related
potentiometry to provide objective neurophysiological information on how sustainable built
environments affect the emotions and productivity of building occupants (Hu et al., 2021).
27
In the future, by using brainwave data in relation to the productivity and mood of the
occupants, designers will also be able to infer building performance in reverse. At the same time,
virtual design technology provides an easy way to make changes to environmental parameters,
significantly reducing upfront costs. The result is that architects will be able to observe
occupants' brainwave responses in a virtual environment to change their designs and thus
determine optimal design solutions for the built environment.

2.4.4 Potential problems with EEG studies
Research has proven that EEG can determine the mood and productivity of the occupants in
a building. Also, with the development of VR technology, designers can use these two
technologies to improve the quality of indoor environments. However, current research does not
delve into which factor or factors have the most significant impact on the building. Thus,
designers are unable to improve the building's indoor environment more effectively.

2.5 Summary
Chapter 2 reviews the literature related to indoor environmental quality and brain signals, as
well as the potential application of EEG in the study of the built environment. This chapter
discusses the importance of IEQ and its impact on human health, productivity, and comfort. At
the same time, it covers the use of federal learning in the building industry, which allows the
development of predictive models for energy consumption and indoor environmental quality
while protecting data privacy. In addition, Chapter 2 discusses the various factors that influence
28
environmental comfort, including temperature, humidity, lighting, and air quality, and their
potential impact on brain signals.
Overall, the field of IEQ and brain signaling has gaps and opportunities for further research.
For example, many studies have explored the specific factors that affect brain signals, and how
user satisfaction can be determined by EEG data. However, these conclusions are not put to good
use. If these relationships can be further utilized, they can be used to predict the user's perception
of the interior environment of the building. It sets the stage for subsequent chapters, which
explore the relationship between IEQ and brain signals in greater depth and present a model for
predicting employee satisfaction based on physiological responses.

29
Chapter 3. Methodology
The relationship between human brain signals and indoor environmental quality was studied.
The study collected various data from volunteers at work, including brain signals, skin
temperature, heart rate, stress, skin humidity, and questionnaires. The data was analyzed to
determine the environmental quality of the work area, with the goal of improving the relatively
poor working conditions. The office building of TK1SC was selected for this study to measure
and investigate. In order to investigate the relationship between human physiological responses
and IEQ, two types of data, physiological signals and environmental parameters, were collected
in this study. Physiological response signals include brain wave data, heart rate, pressure, skin
temperature, and humidity. Environmental parameters include room temperature, light intensity,
acoustics, and air quality.
Furthermore, after the experiment concludes, data processing used time series correlation
analysis and analysis of variance (ANOV A) to establish the relationship between brainwave data
and other indoor environmental parameters. On the one hand, the time series analysis established
the relationship between the real-time brainwave data and the collected indoor environmental
data. On the other hand, if there is a need to investigate the relationship between EEG frequency
intervals and the scores of questionnaires completed by volunteers, ANOV A analysis is the best
choice.
Finally, based on the available data and the associations found, a predictive model was built
using federal learning. The prediction model can predict the actual temperature of the user's
surroundings based on their EEG data from the past few minutes. As a result, more accurate data
30
was available to allow managers to better improve the indoor environment. The flow overview of
the experiment is shown in Figure 3.1.

Figure 3.1 Workflow overview

Figure 3.2 shows the office floor plan of TK1SC. The office is divided into two zones
according to the air conditioning system in the building. Comparing the two areas, we found that
Zone 1 has better light conditions, while Zone 2 has a better view out the window. Meanwhile,
both areas have seats near the window and away from the window. Therefore, a total of four
different areas can be selected for data collection. In order to obtain more comprehensive data,
volunteers will randomly select three different areas to work on.

Figure 3.2 TK1SC floor plan
31

3.1 Experimental equipment
To better analyze the relationship between indoor environmental parameters and human
physiological responses, it is necessary to use a wide range of equipment. This project utilized
six different devices to collect data on various indoor environmental parameters and
physiological responses. The indoor environmental parameters include a light intensity meter, a
temperature, humidity, and carbon dioxide monitor, and a sound level meter. In addition, a
brainwave collection device and two smartwatches were used to collect physiological response
data, including skin temperature, heart rate, heart rate variability, and electrical skin activity.
The use of this diverse set of devices allowed for gaining a more comprehensive
understanding of the relationship between indoor environmental parameters and physiological
responses. Light intensity meters provided information on indoor ambient lighting conditions,
while temperature, humidity, and carbon dioxide monitors provided an understanding of air
quality. Sound level meters helped assess noise levels in the indoor environment, while
brainwave collection devices and smartwatches provided physiological data on participants.
The combination of so many devices were able to help collect a large amount of data on
indoor environmental parameters and human physiological responses, which was crucial for
statistical analyses conducted later in the study.

32
3.1.1 OMEGA HHLM-1 Digital Lightmeter
To measure the light intensity in the office, this experiment used the OMEGA HHLM-1
digital light meter, as shown in Figure 3.3. This portable, easy-to-use 3½-digit, compact digital
light meter is designed for simple, one-handed operation. The light meter measures in lux and
has a backlit LCD display for easy reading. The device's "range" button allows the user to select
the desired lux range for more accurate measurements.

Figure 3.3 OMEGA HHLM-1 Digital Lightmeter

During the study, volunteers recorded room lighting conditions each time they were
surveyed, allowing us to collect data on changes in light intensity throughout the day. It can be
said that the OMEGA HHLM-1 digital light meter provides a reliable and easy-to-use tool for
measuring light intensity in the office environment. The data it collects is essential for analyzing
the relationship between lighting conditions and physiological responses.

3.1.2 HOBO MX1102 carbon dioxide logger
The Onset HOBO MX1102 carbon dioxide logger, shown in Figure 3.4, was used in the
experiment to measure and record carbon dioxide levels in the building environment. This device
makes it easy to obtain accurate measurements of CO2 levels in non-condensing environments in
33
the range of 0-5000 ppm. HOBO can also collect data on the dew point, relative humidity, and
temperature, making it a versatile tool for measuring indoor environmental parameters.

Figure 3.4 HOBO MX1102 carbon dioxide logger

At the same time, users can use a cell phone or tablet to access data from the HOBO within
100 feet via a free app, providing easy access to real-time environmental data. In addition, the
MX1102 has a USB port that allows users to use software to control the device and export data
directly from their computer.
During the volunteers' work, the HOBO MX1102 was placed next to them to obtain data on
temperature, CO2 concentration, and relative humidity. These data were collected at one-minute
intervals, providing a comprehensive data set. It can be said that in studies of indoor
environmental quality and human physiological responses, the Onset HOBO MX1102 has
proven to be a reliable and versatile tool for measuring and recording carbon dioxide levels and
other environmental parameters.

3.1.3 PCE-SDL 1
Figure 3.5 depicts the PCE-SDL 1, a sound-level data logger that measures and records a
wide range of environmental sounds, making it ideal for quality control in noise engineering. The
device is versatile and can be used in factories, schools, offices, major traffic routes, and homes.
34
Users can select either the normal or peak mode of the PCE-SDL 1. In normal mode, the device
captures 20 data points (one data point every 50ms) and records the average of these 20 data
points. In peak mode, the device records data continuously at 50ms intervals, with peaks
indicating the maximum and minimum values. This data can be stored or recorded by the device
in real time. In this project, the sound data is recorded in real time using the normal mode, and
the recording is done at ten-second intervals.

Figure 3.5 PCE-SDL 1

3.1.4 Emotiv Insight 2.0
The Emotiv Insight, shown in Figure 3.6, is a consumer EEG device that measures all
cortical lobe activity, making it unique in its class. The device's polymer sensor technology
provides excellent conductivity with minimal setup, and its hydrophilic material reduces the need
for conductive fluids. Meanwhile, Insight provides users with a range of data types, including
raw EEG, mental commands, and performance metrics such as stress, engagement, interest,
relaxation, focus, arousal, and even facial expressions.
35

Figure 3.6 Emotiv Insight 2.0

These metrics allow users to better understand their performance. For example, they can
determine how long they sustain their attention during the day and regulate their stress levels. In
addition, users can compare their attraction to different stimuli, allowing them to optimize their
performance or modify their environment to create a more favorable atmosphere. The brainwave
data collected from volunteers in this study is a key part of the subsequent data analysis. It will
be used to establish associations with other indoor environmental parameters. Therefore,
comprehensive and accurate EEG data will help identify any deficiencies in the quality of the
office indoor environment and facilitate necessary improvements.

3.1.5 Empatica Embrace 2
The Empatica Embrace 2, depicted in Figure 3.7, is the only FDA-approved wrist-worn
device for the treatment of epilepsy. The device is designed to provide continuous safety and
comfort to the user and is also intended to help people with epilepsy in emergency situations.
Meanwhile, Embrace 2 has a battery life of 48 hours or more, and its quick charge feature can
provide a full day of power to the device in as little as half an hour of charging time.
36

Figure 3.7 Empatica Embrace 2

In addition, Embrace 2 weighs only 13 grams and is waterproof to a depth of 1 meter. When
wearing the device, users can connect to their phones via Bluetooth to transfer data. Finally,
Embrace 2 has four powerful sensors: skin electrical activity, temperature, accelerometer, and
gyroscope. Specific details of these sensors are shown in Table 3.1.

Table 3.1 Sensor Information

3.1.6 Garmin Vivosmart 3
As shown in Figure 3.8, Vivosmart 3 is a smartwatch designed for heart rate monitoring. It
tracks the user's health by continuously monitoring their heart rate throughout the day and
features all-day pressure tracking and a relief-based breathing timer. By tracking the user's heart
rate variability (HRV), the device calculates and displays their stress levels. As a result, users can
easily access their stress levels and other data directly on the device. If the user's stress level
37
rises, indicating that they are becoming more stressed, the Vivosmart 3 will notify them so they
can take steps to relieve their stress.

Figure 3.8 Vivosmart 3

3.2 Experiment procedure
The aim of the experiment was to collect signals of physiological responses and parameters
related to the quality of the indoor environment. The volunteers were all USC students, and ten
people participated. Each participant was in good physical condition and had no specific health
problems. Participants were asked to provide their basic information, such as age, gender, and
name. The specific timeline used is shown in Figure 3.9.

Figure 3.9 Timeline

38
For collecting data on brain signals under different indoor air quality, participants will work
for one hour at each of the three different seats in the company. V olunteers will be asked to take
three surveys during the work period at each seat. The surveys will be administered at the
beginning, middle, and end. The survey content is shown in Table 3.2, which mainly includes the
volunteers' satisfaction with the overall IEQ, temperature, sound environment, light conditions,
indoor air quality, and spatial quality of the seats. At the same time, the questionnaire involves
understanding the quality of the volunteers' work during this period. While the volunteers are
doing the survey, the values of light intensity are also recorded.

Table 3.2 Occupant Satisfaction Questionnaire

Two types of data were also collected - physiological response data and indoor
environmental quality parameters. V olunteers worked as usual during the experiment, and their
physiological response data were recorded. The data recording took about three hours for each
person, and the whole test took no more than four hours. In addition, light intensity values were
recorded during the investigation. The aim was to obtain more comprehensive data and explore
the relationship between physiological responses and indoor environmental quality parameters.
Question Very unsatisfactory Neutral Very satisfactory
How satisfied are you with the overall indoor
environmental quality of your workstation?

How satisfied are you with the thermal?

How satisfied are you with the acoustic?

How satisfied are you with the lighting at your
workstation?

How satisfied are you with the indoor air quality?

How satisfied are you with the spatial quality?

How is your overall productivity at the
workstation?

39
3.3 Data processing
In order to analyze the data, we utilized several statistical software programs, including
Minitab, SPSS, and Python. Minitab and SPSS were used for descriptive statistics, hypothesis
testing, and regression analysis, while Python was used for data cleaning and preparation,
statistical modeling, and visualization. At the same time, Excel was used to organize data and
perform some simple calculations.
To begin the data analysis, we used the corresponding formula in Excel to convert the
timestamps of different digits to US local time. This step was crucial to ensure that all data points
were consistent with the local time zone. Additionally, we utilized the formula of offsetting
average to unify the data of different time intervals into a group of data per minute. This allowed
us to compare the data more efficiently and to identify any patterns that may exist in the data.
Next, we imported all data from the same volunteer, including physiological and
environmental data, into an Excel file, as shown in Table 3.3. By compiling the data in this way,
we were able to organize it more efficiently and analyze it more systematically. This step was
necessary in order to identify any potential relationships or correlations between the different
types of data.

Table 3.3 Data sample from the first volunteer
40

Overall, these initial data preparation steps were critical in ensuring that the data was in a
suitable format for analysis. By unifying the data into a consistent time interval and organizing it
by volunteer, we were able to conduct a more comprehensive and meaningful analysis of the
data.
The data analysis section is divided into two parts, each utilizing different statistical
techniques. The first part focuses on a time series correlation analysis between EEG and real-
time environmental data using SPSS. This analysis will help identify any potential relationships
between the physiological responses of participants and the environmental factors that they are
experiencing. The results will provide insights into how different environmental factors, such as
temperature, carbon dioxide concentration, and noise, affect brain activity and cognitive
performance.
The second part of the analysis will be an ANOV A comparing the results of the
questionnaire and the EEG data using Minitab. This will allow us to examine how subjective
perceptions of the indoor environment relate to the physiological responses of participants. By
comparing these two types of data, we can gain a more complete understanding of how
environmental factors impact cognitive performance and well-being.
Finally, to build the prediction models, we will use RStudio, an integrated development
environment for the R programming language. Both models will use volunteers' real-time
brainwave data to predict their satisfaction with the indoor environment.
41
The first model integrates data from all volunteers and trains the model on this combined
dataset. This trained model can then make satisfaction predictions based on anyone's data.
To predict customer satisfaction, two models were developed using decision trees. The first
model uses EEG data and satisfaction questionnaires for prediction, while the second model adds
IEQ parameters to it. An accuracy comparison of the two models will be made to determine
which one performs better.

3.4 Summary
Chapter 3 outlines the various steps of this study and provides the general framework of the
study. This chapter first explains in detail the selection of the experimental equipment and the
data acquisition process. It then describes the data preprocessing steps, including the conversion
of timestamps to local time and the unification of the data into consistent time intervals. Finally,
the experiment also provides a questionnaire for volunteers, which makes the organization and
analysis of the data more efficient.
Although this chapter only briefly lists the methods that will be used to analyze the data, it
provides an essential overview of the steps involved in preparing the data for analysis. The
specific analysis process is detailed in Chapter 4, which focuses on the results of statistical
analyses using a variety of software, including Minitab, SPSS, Python, and RStudio. In
summary, the steps outlined in this chapter are critical to setting up the study and preparing the
data for analysis, which allows for a more comprehensive and meaningful interpretation of the
results.
42
Chapter 4. Data analysis and results
In order to analyze the relationship between EEG data and indoor environment more
comprehensively, this section will use two analysis methods to establish the relationship between
the two, namely time series lag analysis and ANOV A.
On the one hand, time series lag analysis is a standard statistical method used to investigate
the association between two or more time series variables. In this case, the EEG data,
temperature, CO2 concentration, and background noise data were collected at the same time and
were correlated with each other. By analyzing the time lags between these variables, it is helpful
to identify the factors that have a significant impact on the EEG data. For example, if it is found
that EEG activity may decrease after an increase in CO2 concentration, yet there is a 5-minute
time lag between this, it is easy to conclude that a change in CO2 concentration will have a
significant effect on the brainwave data after 5 minutes.
On the other hand, ANOVA is a statistical method used to determine if there is a significant
difference between the means of two or more groups. In this case, using ANOV A, it is possible to
establish the relationship between the EEG data and the scores of each question in the adjustment
questionnaire completed by the volunteers. By performing an ANOV A, researchers can
determine which factors in the indoor environment are significantly associated with the EEG
data. For example, if the results of the experiment found that the volunteers' scores of satisfaction
with the sound environment had a significant effect on the EEG data. It would be possible to
conclude that the sound environment is an important factor in optimizing the indoor environment
to improve cognitive performance.
43
The experiment could be concluded by combining the results of both analysis methods and
analyzing them together, which would provide a more comprehensive understanding of the
relationship between brainwave data and indoor environment quality. At the same time, the
results of the experiment will be able to identify the key factors that affect the employees' work
status and make informed recommendations for optimizing the indoor environment in the
workplace.

4.1 Time-lagged cross correlation analysis
Time lag cross correlation is a method used to determine the time delay between two signals
or time series. As Figure 4.1 shown, it is evident that the mutual correlation between the red and
blue series can be obtained. The green curve below is obtained if the offset is visualized with the
mutual correlation data simultaneously. The peak of this curve indicates that the two-time series
have the most significant similarity at a specific offset value.
44

Figure 4.1 Time lag cross correlation visualization model

This method is helpful in analyzing the relationship between two signals with time-varying
patterns, such as EEG signals and indoor environmental parameters. By shifting one signal and
comparing it to another, the correlation between two signals with different time lags or offsets
can be calculated. In this investigation, time-lag cross correlations will be used to analyze the
relationship between EEG data and indoor environmental parameters such as temperature, CO2
concentration, and background acoustics. By calculating the correlation of these signals at
different time lags, researchers can determine the time delay between changes in the indoor
environment and changes in the volunteers' brainwave patterns. This information can help
determine which environmental factors have the most critical effect on the volunteers' cognitive
performance.
45
The correlation coefficient measures the relationship between two variables and can be a
value between -1 and 1. A positive value indicates a positive correlation, where variables change
in the same direction, such as Y getting bigger as X gets bigger. A negative value indicates a
negative correlation, where variables change in opposite directions, such as Y getting smaller and
X getting larger. The closer the absolute value is to 1, the stronger the relationship between
variables, and the closer it is to 0, the weaker the relationship between variables.
There are different types of correlation coefficients, of which the Pearson correlation
coefficient is the most common. Other types include the Spearman rank correlation coefficient
and the Kendall rank correlation coefficient. The Pearson correlation coefficient assumes that the
data are normally distributed and that the sample size is large enough. In contrast, the Spearman
correlation coefficient is not subject to these assumptions and applies to non-linear relationships
and outliers.
In this experiment, the Spearman correlation coefficient will be used because it is more
suitable for analyzing the relationship between EEG data and indoor environmental parameters.
It only considers the rank of the corresponding value of each variable and is not affected by the
data distribution.

4.1.1 Participant 1
The first volunteer participated in the experiment on November 7. First, after importing the
data into SPSS, "Bivariate Correlation" was selected in the "Analysis" function bar. Second, in
the "Bivariate Correlation" interface, six different frequencies of EEG signals and three indoor
46
environmental parameters were selected as variables, and all data were included. Finally, in the
"correlation coefficient" section, Spearman was selected, and a two-tailed significance test was
used. Figure 4.2 shows the setup interface for the final data analysis.

Figure 4.2 Time lag cross correlation analysis setup diagram

Table 4.1 presents the correlation results when the time lag is set to 0, indicating that the data
are correlated in real time. The yellow data highlighted in the table represent the most significant
correlations. The data shows that the EEG signals in the frequency bands 4-8 Hz, 8-12 Hz, and
12-18 Hz correlate most strongly with room temperature and acoustic level in the absence of a
time lag. In addition, four frequency bands, 12-18 Hz, 18-25 Hz, 25-32 Hz, and 32-40 Hz,
exhibit significant correlations with indoor CO2 concentration. The correlation coefficients
between room temperature and the three EEG bands were 0.270, 0.388, and 0.376, while the
correlation coefficients between sound level and the three EEG bands were 0.428, 0.363, and
0.297. Finally, the correlation coefficients between the four EEG bands that are significantly
correlated with indoor carbon dioxide concentration are 0.247, 0.212, 0.152, and 0.176.
47

Table 4.1 Cross-correlation analysis without time lag

When the time lag is set to one minute, which reflects the effect of real-time changes in
indoor environmental factors such as temperature, CO2 concentration, and sound level on the
EEG signal after one minute. The results of the analysis are summarized in Table 4.2. Similar to
the previous analysis, EEG signals in the 4-8 Hz, 8-12 Hz, and 12-18 Hz bands remain
significantly correlated with room temperature with correlation coefficients of 0.265, 0.383, and
0.364, respectively. At the same time, the EEG signals in the bands 12-18 Hz, 18-25 Hz, 25-32
48
Hz, and 32-40 Hz are strongly correlated with the CO2 concentration and have correlation
coefficients of 0.246, 0.234, 0.178, and 0.184. Finally, in addition to the three previously
correlated frequencies, the analysis also reveals a significant correlation between sound level and
EEG in the frequency range of 32-40 Hz, suggesting that an increase in background sound level
may negatively affect EEG in this frequency band in addition to the previously identified
frequency band.

Table 4.2 Cross-correlation analysis with 1-minute lag
49

Table 4.3 shows the correlation analysis data when the time lag is set to 2 min. The results
indicate that room temperature remains significantly correlated with the EEG signals at 4-8 Hz,
8-12 Hz, and 12-18 Hz with correlation coefficients of 0.137, 0.207, and 0.245. Meanwhile, the
CO2 concentration is significantly correlated only with the EEG signals at 12-18 Hz and 18-25
Hz, with correlation coefficients of 0.163 and 0.145. However, the correlation with the other two
frequency bands disappears. Finally, in addition to significant positive correlations with the
frequencies 4-8 Hz, 8-12 Hz, and 12-18 Hz, the sound level also has significant negative
correlations with the EEG waves at 25-32 Hz and 32-40 Hz.
50

Table 4.3 Cross-correlation analysis with 2-minute lag

If the time lag is set to 3 minutes, the results are shown in Table 4.4. As before, room
temperature is only significantly correlated with the first three EEG frequencies, with
correlations of 0.265, 0.381, and 0.354, respectively. In addition, carbon dioxide concentration is
associated with EEG signals at 12-18 Hz, 18-25 Hz, 25-32 Hz, and 32-40 Hz. Finally, acoustic
levels are related to all frequencies except 18-25 Hz and 25-32 Hz, which are not significantly
51
associated. However, the correlation coefficient is -0.214 for the frequency 32-40 Hz, which is
negative, and positive for the other frequencies.

Table 4.4 Cross-correlation analysis with 3-minute lag

When the time lag is set to 4 minutes, the time lag cross-correlation analysis produces
relatively significant changes, and the results are shown in Table 4.5. First, the indoor
temperature parameter is correlated with the EEG signals at only two frequencies, 8-12 Hz and
12-18 Hz, and the correlation coefficients are 0.204 and 0.243. Simultaneously, the CO2
52
concentration is associated with the 12-18 Hz and 18-25 Hz frequencies. Finally, four
frequencies, 4-8 Hz, 8-12 Hz, 25-32 Hz, and 32-40 Hz, are significantly correlated with the level
of acoustics.

Table 4.5 Cross-correlation analysis with 4-minute lag

The research conducted a time-lagged cross correlation analysis with a maximum lag time of
5 minutes, and the results obtained for the maximum lag time are listed in Table 4.6. The analysis
53
indicates that there is still a significant correlation between room temperature and brain waves at
the frequencies of 4-8 Hz, 8-12 Hz, and 12-18 Hz. In addition, carbon dioxide concentration is
found to be relevant for all four frequencies except 4-8 Hz and 8-12 Hz. Finally, the sound level
does not significantly correlate with the EEG signal only at 18-25 Hz, but it shows significant
negative correlations with the frequencies 25-32 Hz and 32-40 Hz.

Table 4.6 Cross-correlation analysis with 5-minute lag

54
Table 4.7 presents the results of the time-lagged cross-correlation analysis between indoor
environmental factors (room temperature, CO2 concentration, sound level) and EEG frequencies
for the first participant. The values in the table indicate the maximum correlation coefficients
between EEG frequency and environmental factors, with the time lags in parentheses. The table
determines after how long an interval each environmental factor had the most significant effect
on the different EEG frequencies of participant 1. For example, there is a significant correlation
between the 4-8 Hz brain waves of person one and both room temperature and sound level,
indicating that changes in these factors had the most excellent effect on the subject's brain waves.
The correlation coefficients between these factors and 4-8 Hz EEG waves decreased over time,
which suggests that the effects of these factors were direct rather than lasting. In addition, for
participant 1, changes in CO2 concentration had the most pronounced effect on his 18-25 Hz
brainwaves after a three-minute time lag.

Table 4.7 Integration of time lag cross correlation analysis results

4.2 ANOVA analysis
Analysis of variance using Minitab on EEG signals and questionnaire results provided a
more comprehensive understanding of the data. Specifically, the experiment will analyze each
participant's EEG data for the last 20 minutes in each seat, along with their responses to the final
questionnaire. This approach ensured that any potential effects of the previous seat's indoor
55
environment on the participants' work status were avoided, as the volunteers were working in
their current seat for at least 40 minutes.
In addition, ANOV A was chosen because it possesses two potential advantages. On the one
hand, it allows for a comprehensive examination of the data set, including trends and patterns
that may not be immediately apparent. With non-real-time data, researchers can scrutinize the
data and make comparisons across groups or conditions to identify potential correlations or
causal relationships.
On the other hand, it allows the identification of significant differences between groups or
conditions, even if these differences are relatively small. This is important because small
differences may be overlooked if only simple mean comparisons are made.
In order to create a visual representation of the data, One Y and With Group in Interval Plot
were selected in the Graph menu bar. Then, in Graph variables select six different frequencies of
EEG signals and select seven questions about satisfaction in order in Categorical variables for
grouping. The specific settings are shown in Figure 4.3. Finally, select MEANS in the data label
to make the generated graphs show the data averages.
56

Figure 4.3 Minitab data analysis settings

4.2.1 Overall IEQ satisfaction and EEG data
The results of the ANOV A between participant 1's overall indoor environmental satisfaction
survey scores and EEG data for three different seats are shown in Figure 4.4. As participant 1
only provided two overall indoor environmental satisfaction scores for the three seats, the
ANOV A analysis produced only two regular distribution intervals for the scores.
57

Figure 4.4 ANOV A analysis of overall indoor environmental satisfaction and EEG data

The figure shows that significant differences exist in the EEG distribution intervals for
different scores across three frequencies: 4-8 Hz, 8-12 Hz, and 18-25 Hz. The differences in
brain signal frequencies for 12-18 Hz and 25-32 Hz are also significant, with p-values less than
58
0.05. However, the difference in brain signal frequencies corresponding to the two fractions in
the graph for the frequency 32-40 Hz is small, and the p-value is greater than 0.1, suggesting that
this set of data is not useful for further analysis.

4.2.2 Thermal satisfaction and EEG data
The ANOVA was used to analyze the thermal satisfaction scores of Participant 1 at three
different workstations and the corresponding EEG data, as shown in Figure 4.5. The results
indicate that the volunteers' perception of temperature varied significantly between workstations,
highlighting the importance of controlling room temperature in this context.
59

Figure 4.5 ANOV A analysis of thermal satisfaction and EEG data

ANOV A revealed significant differences in the mean values of the EEG signals
corresponding to different thermal satisfaction scores within each frequency interval. The p-
values obtained from the ANOV A tests were all less than 0.05, indicating significant differences
in the frequency ranges associated with each score.
60
4.2.3 Background acoustic satisfaction and EEG data
The results of the analysis of Participant 1's satisfaction with background acoustic and EEG
data in the office are presented in Figure 4.6. Since there is no sound insulation between the seats
in the office, the noise levels in the same areas do not vary significantly, resulting in only two
possible scores for the volunteers.

Figure 4.6 ANOV A analysis of background acoustic satisfaction and EEG data

61
In addition, the graph shows that the EEG signals in all frequency ranges, except for the 8-
12 Hz range, have completely different regular distribution intervals for the two scores.
However, for the 8-12 Hz range, there is an overlap between the EEG frequency intervals
corresponding to the two scores. The specific value also shows that the p-value between the two
data sets in this interval is 0.791. This value is significantly greater than the prescribed 0.05. It
also demonstrates that there is no significant difference between the two data sets. Therefore,
except for the 8-12 Hz range, the frequencies corresponding to different scores show significant
differences in the data covariates.

4.2.4 Illumination intensity satisfaction and EEG data
Figure 4.7 shows the results of the ANOV A analysis of the light intensity satisfaction with
the corresponding EEG data at different workstations for participant 1. It can be seen from the
figure that the satisfaction of volunteers differed significantly across locations due to the
difference in illumination intensity.
62

Figure 4.7 ANOV A analysis of illumination intensity satisfaction and EEG data

In the 4- 8hz frequency interval plot, there is an overlap between the EEG distribution plot 4
and 7 points. Similarly, the distribution of intervals corresponding to scores 3 and 5 in the 12- 18
Hz plot showed some overlap. Despite these overlaps, the p-values obtained after ANOV A
analysis of the EEG data and corresponding scores for all frequency intervals were less than
0.001, indicating a statistical difference between the means of the EEG frequencies for each
score.
63
4.2.5 Indoor air quality satisfaction and EEG data
As the indoor air quality did not vary much within the same areas of the company, only two
scores were reported. Figure 4.8 presents the analysis of Participant 1's satisfaction with indoor
air quality and the corresponding EEG data. The Interval Plots show that the EEG signals at 4-8
Hz and 18-25 Hz frequencies have overlapping parts in the normal distribution interval of the
two scores.

Figure 4.8 ANOV A analysis of IAQ and EEG data

64
The analysis reveals that the overlap between the two intervals in the 4-8 Hz Interval Plot is
smaller, and the obtained p-value of 0.019 is less than the required 0.05, indicating variability in
the distribution interval corresponding to the two scores at this frequency. However, the part of
the overlap between the two intervals in the Interval Plot of 18-25 Hz is larger, and the P-value
of 0.5 is much larger than the required 0.05, so it has no subsequent reference significance.
Overall, the analysis showed variability in the distribution intervals corresponding to
different scores for all EEG signal frequencies, except for the 18-25 Hz frequency.

4.2.6 Spatial quality satisfaction and EEG data
The satisfaction survey of space quality includes an essential factor of the view near the seat.
For example, if the workstation is near a window and the employee can see a beautiful view
outside the window, then the score of this item will be higher. The result of the ANOV A analysis
is shown in Figure 4.9.
65

Figure 4.9 ANOV A analysis of spatial quality satisfaction and EEG data

According to the results obtained from the ANOV A, it can be found that all the Interval Plot
graphs have variability in the distribution intervals of the brain waves corresponding to different
scores because all the p-values are less than 0.001, which means much less than the required
0.05.

66
4.2.7 Work efficiency satisfaction and EEG data
The ANOVA analysis results of work efficiency and EEG data are presented in Figure 4.10.
The findings suggest that there is variability in the distribution intervals for all frequencies
except in the Interval Plot of 4-8 Hz and 18-25 Hz charts, where an overlap exists between the
normal distribution intervals corresponding to the two scores.

Figure 4.10 ANOV A analysis of work efficiency satisfaction and EEG data

67
The small overlap in the Interval Plot of the 4-8 Hz graph and the confirmed p-value of
0.019, which is less than 0.05, implies that there is variability between the two. However, the
overlap in the 18-25 Hz plot is too large, and the P-value of 0.5 suggests that it is not a reliable
reference.

4.3 Results
Both the time-lagged cross-correlation analysis and the ANOVA were performed by first
analyzing each individual's EEG signal data with the corresponding indoor environmental data
and questionnaire data. The results obtained from the analysis were valid for the individual.
However, aggregating data from multiple individuals can provide a more comprehensive
understanding of the relationship between indoor environmental factors, subjective perceptions,
and EEG signals. By combining data from multiple individuals, it is possible to identify common
patterns and trends across the group.

4.3.1 Summary of time lag cross correlation analysis
The time interval between each volunteer's own response to changes in the indoor
environment varied. In Table 4.8, the EEG signals of each individual are presented with
significant correlation with different indoor environment parameters, and the time corresponding
to the maximum correlation coefficient is shown. The table includes different time lags, different
EEG frequency intervals, and three indoor environmental parameters.
68

Table 4.8 Correlation analysis between EEG and environmental data

First, regarding indoor temperature, there are two cases of correlation coefficients. On the
one hand, the indoor temperature is positively correlated with the EEG signal level at frequencies
below 18 Hz and negatively correlated with frequencies above 25 Hz. At the same time, the
effect on the EEG waves of 18-25Hz frequencies was smaller. In addition, when the temperature
exceeds the comfort zone of the human body, if the temperature continues to rise and fall, it will
have a negative impact on human physiology.
Second, the effect of CO2 concentration on brain activity is relatively small, as people do
not perceive small changes in indoor CO2 concentration directly from brainwave data. However,
CO2 concentration is positively correlated with brain waves at 4-8 Hz, 8-12 Hz, 12-18 Hz, and
18-25 Hz. In addition, it is negatively correlated with EEG signals at two frequencies above 25
Hz.
Finally, the effect of sound on brain wave frequencies is positively correlated. The office
environment in this study had low noise levels, but if a room is too quiet, it may not promote
staff productivity. Therefore, appropriate sound can positively impact people's brain activity.
69
Table 4.9 displays the number of individuals who exhibited the maximum correlation
coefficient at varying time lags during changes in indoor environmental parameters. For instance,
when the room temperature was altered, three out of nine volunteers displayed the highest
correlation coefficient after five minutes for 4-8 Hz frequency EEG waves. The table shows that
there were 3,4, 3, 2, 3, and 4 individuals respectively, whose time lags applied to the relationship
between the EEG signal and the room temperature at all frequencies. Therefore, it is obvious to
conclude that room temperature changes produce the greatest effect on human physiological
responses after 5 minutes.

Table 4.9 Number of people applicable at different time intervals

In addition, the time interval distribution of the response made by people's EEG signals is
complex when the CO2 concentration changes. For frequencies below 25 Hz, the volunteers
physiologically responded to changes in CO2 concentration between 0 and 1 minute. Meanwhile,
for frequencies above 25 Hz, the response time was about 4 minutes.
Finally, it is clear that most of the volunteers can be found to have changes in the EEG signal
at the same time as the change in sound decibels. The numbers of people were 5, 3, 3, 2, and 2
affected by the real-time sound changes. Therefore, the physiological effects of sound level
changes on the staff were in real time.

70
4.3.2 Summary of ANOVA analysis
Table 4.10 presents the results of the ANOV A analysis, which indicate a significant
association between the volunteers' satisfaction ratings and EEG signals for heat, sound level,
light intensity, and spatial quality. In the table, for each frequency of the EEG signal, more than
half of the volunteers' EEG data proved to be associated with their questionnaire scores. These
indoor parameters play a crucial role in the overall indoor environment, indoor air quality, and
work productivity. Therefore, it is more likely to be perceived visually by employees.

Table 4.10 ANOV A analysis summary

4.4 Summary
Both analysis methods reveal a strong correlation between real-time data, such as EEG
signals, and thermal and acoustic levels. On the one hand, it takes about 5 minutes for people to
change and respond to the indoor temperature. On the other hand, people's physiological
response rate to background noise can be considered as real-time.
At the same time, the changes in EEG data are closely related to some indoor environmental
factors that are more intuitively perceived by employees. However, for subjective issues such as
overall satisfaction and work efficiency, the relationship between the scores given by employees
and their actual EEG signals is not clear, making it difficult to identify patterns.

71
Chapter 5. Prediction model
As indoor environmental quality becomes more and more important in the workplace, there
is a greater need for an accurate and comprehensive predictive model to help improve IEQ in the
office. Based on research into machine learning and predictive models, the use of federated
learning would be a promising approach. It is a machine learning technique, but it allows data
training across devices or servers without the need for centralized data storage. At the same time,
by using federated learning, customers can train the model on their own mobile devices with data
and eventually upload the obtained results to the master server. The result is that using federated
learning can protect the user's privacy and maintain data security while keeping the amount of
data available.
Federated learning can be used to develop a prediction model that leverages physiological
response data to predict employee satisfaction with indoor environmental quality. For example,
real-time data from sensors can be collected to predict whether building temperatures will be
satisfactory for employees, allowing for the optimization of building systems to provide a more
comfortable and healthy work environment. By incorporating occupant physiological data, this
approach presents an exciting opportunity to improve building efficiency, sustainability, and
overall occupant satisfaction with IEQ.

5.1 Establishment of support vector machine model
In order to be capable of finding a suitable model for building prediction models based on
federal learning methods, it is necessary to select some base algorithms suitable for this data and
72
to compare them. In the context of predicting heat satisfaction using six-channel EEG data and
seven-point scale survey data, the support vector machine (SVM) algorithm can be used to build
a prediction model. The SVM algorithm is a powerful machine learning method that works by
finding the best decision boundary between two classes of data. In this case, the decision
boundary divides the data into two categories: satisfied and dissatisfied with the indoor
temperature.

5.1.1 Predictive model algorithm selection
Once the input data has been processed, the SVM algorithm can construct a predictive model
by finding a decision boundary that maximizes the margin between the closest data points in
each class, called the support vector. The SVM algorithm is well suited for processing high-
dimensional data such as EEG signals because it can identify complex relationships between
input features and target variables.
However, it is essential to notice when using support vector machine models to build
predictive models that their accuracy will depend on the quality and representativeness of the
input data. The dataset which is used to train and test the SVM model should be large and diverse
enough to capture individual differences in thermal comfort perception. It is equally inevitable to
be aware of any potential biases or limitations in the data, such as skewed distributions of survey
scores or sampling rate variations in EEG data.
The trained SVM model can be used to predict thermal satisfaction scores for new data
points. The model can be applied to various settings, such as building design or HV AC system
73
control, to improve thermal comfort decisions. However, when using the model, it is necessary to
be aware that there are potential challenges in deploying it in real-world settings, such as the
need for continuous data collection and individual differences in thermal comfort that may affect
the accuracy of the model.
Overall, the SVM algorithm provides a powerful tool for thermal satisfaction prediction
based on EEG and survey data inputs. However, what is most significant is the need to determine
whether the quality and representativeness of the data used to train and test the model meets the
requirements, as well as the potential limitations of applying the model in real-world settings.

5.1.2 Algorithm Explanation
The Coding Script of the model that uses the SVM algorithm to predict employee
satisfaction with indoor sound levels is shown in Figure 5.1. The critical parts of the Coding
Script will be explained.
74

Figure 5.1 The Coding Script of the SVM algorithm

Figure 5.2 shows the code of data selection SVM algorithm for data selection. For example,
EEG signal frequencies from 4-8Hz to 32-40Hz and employee satisfaction scores with sound
levels are selected in the figure as the base data for this model.
75

Figure 5.2 Data Selection

The code in Figure 5.3 splits the data in the prediction model into a training set and a test set.
For example, in the sample function section, it will select 70% of all the data as the training set
for changing the model. Meanwhile, the remaining 30% of the data will be used as the test set to
test the accuracy of the model. This step is important to ensure that the model can accurately
predict new, unseen data.

Figure 5.3 Data splitting

The code in Figure 5.4 implies using the training dataset for SVM model training. The data
set sound is the response variable, and all other variables are predictor variables. At the same
time, the code reveals that the sum function type of the SVM algorithm is a linear function.

Figure 5.4 SVM model training

In order to be able to weigh whether the prediction model can be used in real situations, it is
necessary to check the accuracy of the reformulated model. Figure 5.5 shows the code related to
the accuracy prediction of the SVM model. First, the model makes predictions on the test data
and saves the predicted data into a new vector. Second, the predicted results are compared with
the actual values in the test data to obtain the accuracy of the SVM model. Finally, the resulting
76
model accuracy is presented as a new data frame with predicted and actual values.

Figure 5.5 Accuracy prediction

5.2 Application of decision tree model
Decision tree classification is a powerful supervised learning algorithm that can help create
predictive models based on input data. In this case, EEG and IEQ data can be used as input data
to predict satisfaction with IEQ. The decision tree algorithm works by recursively segmenting
the input data based on the values of the input features until a stopping criterion is reached.
Decision tree construction can be performed in two steps. The first step is decision tree
generation: the process of generating a decision tree from the training sample set. The second
step is decision tree pruning: decision tree pruning is the process of checking, correcting, and
trimming down the decision tree generated in the previous stage, mainly by checking the
preliminary rules generated during decision tree generation with the data in the new sample
dataset (called test dataset) and cutting out those branches that affect the accuracy of pre-
balancing.
There are several benefits to using decision trees for volunteer satisfaction prediction. On the
one hand, one of the benefits of using the model for the task is that it is easy to interpret and
visualize. Decision trees can be plotted to show the decision rules used for prediction, which can
help in understanding the relationship between input features and output variables. Another
77
advantage is that decision trees can handle categorical and continuous input features, making
them suitable for processing EEG and IEQ data.
Therefore, decision tree classification is a valuable tool for creating predictive models based
on input data to predict satisfaction with IEQ.

5.2.1 Decision Tree algorithm explanation
The prediction model will be built in Python. First, as shown in Figure 5.6, this is a code
snippet for importing the libraries required for decision tree classification in Python. Pandas is a
library for data processing and analysis in Python, while scikit-learn is a machine learning library
that provides a range of tools for data analysis and predictive modeling. The tree module is used
for decision tree-based models. It also imports the matplotlib.pyplotas library for data
visualization purposes.

Figure 5.6 Library import

This code fragment in Figure 5.7 creates three variables: featres1, x1, and y1, which are used
for decision tree classification. Features1 is a list of strings containing the column names of the
characteristics used for classification. The purpose of this code snippet is to prepare data for the
input decision tree model, where x1 contains the input features and y1 contains the
78
corresponding target variables.

Figure 5.7 Decision tree variables creation

As shown in Figure 5.8, the code creates a decision tree classifier with a maximum depth of
4 and trains it with the data stored in the variables "x1" and "y1". In addition, it plots the decision
tree using the "plot tree" function in the "tree" module and saves it as an image file.

Figure 5.8 Decision tree classifier generation

Finally, the code in Figure 5.9 imports the metrics module from the sklearn library, which
contains functions to calculate various evaluation metrics. It then prints the accuracy score, which
is the percentage of correctly predicted results to the total number of predicted results. y_test and
y_pred1 variables are the true and predicted values.

Figure 5.9 Accuracy value export

5.2.2 Sample decision tree
An example of using decision trees for volunteer satisfaction prediction is shown in Figure
79
5.10. The figure shows the use of participants' EEG data to predict their satisfaction with the
heat. A tree is a flowchart-like structure that depicts all possible outcomes of a series of related
decisions. Every single node in the decision tree represents a decision or a set of decisions, and
the branches emanating from each node represent the possible outcomes of these decisions. In
this decision tree, the nodes segment the data according to different bands and their values, and
the Gini values for each segmentation are shown. The sample and value values show the number
of data points on that node and the number of each class. Also, this figure can be used to gain
more insight into how the decision tree makes predictions. In addition, by visualizing the
decision process in this way, the logic behind the tree predictions can be more easily
communicated to those who may not be familiar with the underlying data or machine learning
techniques.

Figure 5.10 EEG and thermal satisfaction decision tree

The figure illustrates the classification of brain waves in a model containing 81 data entries.
The first row in the figure shows the decision tree, starting with the classification of brain waves
80
from 12-18 Hz. Moving to the second row, we observe the processing of the decision tree in two
cases. For example, when the data is greater than 0.739, the decision tree shows "no" and groups
the data to the right for further classification. Similarly, when the data is less than or equal to
0.739, the decision tree shows "yes" and groups the data to the left for further classification.
Further analysis of the graph reveals that of the 81 data entries, there are two groups of 12-18 Hz
EEG values less than 0.739. This indicates that these two groups are somehow different from the
rest of the data. However, the remaining 79 entries require additional classification. This
indicates that the model is complex and requires further analysis to fully understand the
classification process.

5.3 Reliability test of the prediction model
A validation experiment is necessary to assess the reliability of the prediction model
developed in this study. The experiment was based on the established model to analyze the
operations on other remaining volunteer data. The predicted results of the model were compared
with the actual data to test the accuracy and generalization of the prediction model.
To visualize the performance of the SVM model, it was necessary to create a scatter plot of
predicted versus actual heat satisfaction scores, as shown in Figure 5.11. The scatter plot
provides an easily understandable graphical representation of the predictive power of the model.
The data in the figure shows that the SVM model can be used to predict employee satisfaction
with indoor environmental quality predictions.
81

Figure 5.11 Scatter Plot of Predicted vs. Actual Thermal Satisfaction Scores

However, the actual prediction results showed that the SVM model obtained unsatisfactory
prediction results. The red line in the scatter plot represents the predicted scores of volunteers'
satisfaction with heat. Since the prediction scores are concentrated around five points, they are
not very meaningful for reference. Therefore, a potential solution to improve the accuracy of
predictive models is to collect more data for analysis.
The decision tree model is more precise than the SVM model in classifying scores. This is
because it can recursively segment input data based on the values of input features until a
stopping condition is reached. This makes the decision tree model ideal for handling EEG and
IEQ data, and it can achieve high levels of accuracy. In fact, using only EEG data, the model can
accurately predict thermal and acoustic comfort with an accuracy rate of 84%. When both EEG
82
and IEQ data are used together, the accuracy rate remains high at 80%. These results demonstrate
the potential of the decision tree model in accurately predicting the comfort levels of individuals,
making it a valuable tool for use in various fields.

5.4 Summary
On the one hand, by using support vector machine algorithms, a powerful machine learning
method, a predictive model can be constructed to detect complex relationships between input
features and target variables. However, the effectiveness of the SVM algorithm is affected by the
quality and representativeness of the input data, and there are biases or limitations in the
prediction data obtained in this experiment.
On the other hand, this section also discusses the decision tree algorithm as a practical tool
for designing prediction models based on input data to predict IEQ satisfaction. The decision tree
algorithm recursively segments the input data based on the values of the input features until a
stopping condition is reached. The decision tree is also capable of managing categorical and
continuous input features, making it suitable for processing both EEG and IEQ data.
Overall, by conducting validation experiments to assess the reliability of the created
prediction models, the chapter demonstrates that the decision tree algorithm has more potential to
accurately predict personal comfort.

83
Chapter 6. Conclusion and future work
This section provides a comprehensive summary of the previous chapters, including the pre-
experimental background investigation and the experimental procedures used in this study. At the
same time, this chapter describes the limitations of this study and present potential improvements
for future studies. In addition, it suggests potential improvements to the prediction model based
on the analysis of the results obtained in the previous chapters. Finally, possible future research
directions are proposed to expand the scope of this study and to promote a more in-depth
understanding of the relationship between human physiological responses and IEQ satisfaction.

6.1 Discussion
The value society places on indoor environmental quality has led to a more comprehensive
assessment of it. Federated learning has emerged as a valuable tool for analyzing IEQ and human
physiological responses while also safeguarding the privacy of volunteers.
Based on the existing literature, various factors affecting the indoor environment can have an
impact on the human EEG signal. At the same time, IEQ is important for human health,
productivity, and comfort. Moreover, the use of federal learning in the building industry allows
the development of predictive models for energy consumption and indoor environmental quality
while protecting data privacy. However, additional investigations are necessary to gain a
complete understanding of the relationship between IEQ and brain signals using this new
learning model.
84
Collecting volunteer data is critical for providing the necessary information to the predictive
model. Therefore, experiments were conducted at an innovative engineering company in
Southern California that has implemented a dynamic platform. The experiments collected
physiological signal data and indoor environmental parameters from volunteers, along with their
scores for IEQ satisfaction on a questionnaire.
Data will be analyzed using SPSS and Minitab with time-lagged cross correlation and
ANOV A analysis. The time lag correlation reveals that the EEG signal has the most significant
effect after a 5-minute change in indoor ambient temperature, while changes in background noise
immediately affected the employees' brainwave data. In addition, ANOV A analysis of the
relationship between EEG data and volunteer satisfaction surveys found that changes in EEG
data were closely related to some indoor environmental factors that were more intuitively
perceived by employees. However, the relationship between scores given by employees for
subjective questions, such as overall satisfaction and work efficiency, and the actual EEG signal
was unclear, making it difficult to identify patterns.
As indoor environmental quality becomes increasingly important in the workplace, there is a
greater need for an accurate, comprehensive predictive model to help improve office IEQ.
Federated learning, a machine learning technique that allows data training across devices or
servers without centralized data storage, provides a promising solution. A predictive model that
utilizes physiological response data to predict employee satisfaction with IEQ can be developed
using the data collected from experiments and the associations obtained from data obtained from
data analysis. For example, real-time data from sensors can be collected to predict whether
85
building temperatures will satisfy employees, thus optimizing building systems to provide a
more comfortable and healthy work environment. After the model was built, a validation
experiment was conducted on all volunteer data to assess the reliability of the predictive model
developed in this study. The results demonstrated that the model can be used to predict employee
satisfaction with indoor environmental quality.

6.1.1 Evaluation of the workflow
The entire process of the experiment was carefully considered and planned to ensure that
sufficient data were collected and analyzed. First, during the preparation phase, a total of six
devices were selected to collect human physiological response data and indoor environmental
parameter data. At the same time, volunteers were asked to complete a questionnaire on indoor
environmental satisfaction to collect additional valuable data.
Second, during the experiment, participants were asked to work in different locations for one
hour. However, it was essential to avoid the impact of the previous workstation environment on
the participants' new workspace. To prevent such a situation, volunteers were given sufficient
time to adjust to the new environment after changing locations. This adjustment time was critical
to ensure that the data collected were free of bias and accurately reflected the participants'
responses.
Finally, data analysis and the selection of an appropriate algorithm were essential to creating
a reliable and accurate predictive model. Data analysis was conducted to determine the
association between indoor environmental parameters and human physiological responses. This
86
analysis ensured that only data with significant correlations were used in the predictive model,
resulting in more convincing and reliable results.
To conclude, the careful and meticulous planning of the entire experiment, from data
collection to predictive modeling, ensured that the results obtained from the predictive models
were comprehensive, convincing, and reliable. Thus, the improved predictive models and
assessment methods developed using these results are ready for practical application.

6.2 Research Limitations
Although the study has been carefully planned and executed from preparation to the final
conclusion, there are still some limitations. These limitations can be classified into two types:
limitations of the experimental instrument and inadequate considerations in the experimental
methodology.

6.2.1 Limitations of the tools
While experimental equipment can provide detailed data, it is critical to be conscious of the
limitations of their capabilities. For example, Insight 2.0 is an excellent device for collecting
EEG signals. However, female volunteers with longer hair had difficulty adjusting the device
during the experiment, which resulted in longer preparation times to ensure accurate data
collection. The issue of hair length raises concerns that the wearability of the device may be
compromised, leading to data collection issues.
The other device used in the experiment, HOBO, was expected to collect data on room
87
temperature, humidity, and CO2 concentration simultaneously. However, during the experiment,
the device was unable to collect humidity data. This problem meant that the relationship between
EEG signals and changes in room humidity could not be analyzed during data analysis. This
limitation may have affected the comprehensiveness of the results.

6.2.2 Methodological limitations
If these limitations in the experimental methodology can be addressed, the results of this
experiment have the potential to be improved. First, the experimental equipment is insufficient,
resulting in a limited amount of data collected in a given time period. During data analysis, a
larger sample size would lead to more accurate results.
Second, the scope of data inclusion could be broader. The volunteers for this experiment
were all students, which meant that the age range was limited to about 25 years old. However,
the age range of employees working in the office would be much broader. Therefore, if
volunteers of different age groups could be found, the experiment results would cover a more
comprehensive population. In addition, all data were collected in November, which is the fall
season in Los Angeles. Since the indoor environment varies throughout the year, the
physiological responses of employees to changes in the indoor environment may also be
different. Collecting and analyzing data from different seasons can provide detailed information
on employee usage at each location.
Third, additional software can be used to analyze the data to verify the correctness of the
analysis results. It is also possible to find the most easily understandable results among these
88
results, such as some soft solutions to visualize the data. Visualization of data results can help
data analysts communicate the value of the data more effectively.
Finally, data analysis requires not only the number of data samples but also the accuracy of
the data. Data with low accuracy can easily lead to uninformative or biased final analysis results.
Therefore, when choosing an experimental device, it is important to select one that is more
accurate. the Emotiv Insight 2.0 is a brainwave acquisition device with five sensors, but there is
also an acquisition device with 14 sensors. Comparing the accuracy of the data from these two
devices can help to select a more accurate device and obtain more accurate experimental results.
In addition, it may be beneficial to collect data over a more extended period of time to observe
changes in employee productivity and comfort over time.

6.3 Future work
This experiment has some inherent limitations that can be addressed to improve the study of
this topic. Although the existing experimental results suggest potential, there is still much to
explore. Further research can be conducted by optimizing existing methods or conducting more
in-depth studies. In addition, both short-term and long-term improvements must be considered to
obtain more comprehensive and convincing results.
In the short term, research teams may consider expanding sample sizes, including more
diverse populations, or refining experimental designs. In addition, the use of more advanced
technologies and methods could help collect more accurate and detailed data.
89
In the long term, the research team could explore new methods or theories or even expand
the scope of the study. In addition, a more in-depth study of the link between EEG and indoor
environmental quality is necessary so that a better understanding of the topic can be obtained.

6.3.1 Potential research directions
The relationship between the physiological responses of occupants and the indoor
environment has been studied in office buildings. In the future, however, researchers can expand
their investigations to explore this relationship in different types of buildings. By doing so, they
can gain a more comprehensive understanding of how the indoor environment affects the
physical and mental health of occupants, and they can develop more effective strategies to
optimize the indoor environment in different types of buildings.
There are two promising research directions for this topic. One is the application of artificial
intelligence (AI) and machine learning. These cutting-edge technologies can help researchers
analyze the complex connections between physiological data and environmental factors, and they
can provide predictions of future trends based on these connections.
Federated learning, as a kind of machine learning, can be used to develop a predictive
model, which in this study uses physiological response data to predict employee satisfaction with
indoor environmental quality. For example, real-time data from sensors can be collected to
predict whether building temperatures will satisfy employees, thereby optimizing building
systems to provide a more comfortable and healthy work environment.
90
At the same time, it can be perfectly used to build a predictive model based on occupant
physiological data. The model aims to predict occupant satisfaction with IEQ by analyzing
occupants' physiological responses to various environmental parameters such as temperature,
sound, light, and air quality. The approach provides real-time and objective feedback on IEQ,
enabling building managers to optimize building systems to improve occupant satisfaction and
productivity.
Another way is the potential use of virtual reality (VR) technology in architectural design.
With VR technology developing, designers are now capable of generating virtual building
scenarios and simulating the physiological reactions of occupants to various environmental
factors.
This approach can help designers modify the building in the virtual scene according to the
relationship between the user's physiological responses and the interior environment, saving time
and resources while meeting the user's requirements for the building's interior environment. By
exploring these research directions, we can gain a deeper understanding of the relationship
between occupants and the indoor environment, and we can develop more effective strategies to
optimize the indoor environment for various types of architectural designs.
In summary, the use of artificial intelligence, machine learning, and VR technologies has the
potential to revolutionize the way we assess and improve the quality of indoor environments. By
exploring these research directions and developing accurate and comprehensive predictive
models and assessment methods, it is possible to create healthier and more sustainable built and
working environments that meet the needs of occupants in different types of buildings.
91
6.4 Summary
The study investigated the effects of indoor environmental quality on human physiological
responses. The study, conducted at a Southern California engineering firm, analyzed
physiological signal data, environmental parameters, and IEQ satisfaction scores. Various indoor
environmental factors were found to affect EEG signals, and predictive models could optimize
building systems to create a healthier work environment. Limitations of the study include the
experimental equipment, which could be improved with larger sample sizes, more diverse
populations, and better data collection methods. The chapter also discusses potential research
directions, such as the use of artificial intelligence, machine learning and virtual reality
technologies to assess and improve indoor environments. The research theme has proven to have
the potential to continue in depth.

92
References:
Anonymous. (2022). WHY IEQ MATTERS BEYOND BASIC COMFORT. Air Conditioning,
Heating & Refrigeration News, 275(3), 23.
http://libproxy.usc.edu/login?url=https://www.proquest.com/trade-journals/why-ieq-matters-
beyond-basic-comfort/docview/2624697217/se-2?accountid=14749

Babapour, M., Karlsson, M., & Osvalder, A.-L. (2018). Appropriation of an Activity-based
Flexible Office in daily work 1. In Nordic journal of working life studies (Vol. 8, Issue S3).
https://tidsskrift.dk/njwls/index.
Bower, I. S., Clark, G. M., Tucker, R., Hill, A. T., Lum, J. A. G., Mortimer, M. A., & Enticott, P.
G. (2022). Built environment color modulates autonomic and EEG indices of emotional
response. Psychophysiology, n/a(n/a), e14121.
https://doi.org/https://doi.org/10.1111/psyp.14121

Castaldo, V . L., Pigliautile, I., Rosso, F., Cotana, F., de Giorgio, F., & Pisello, A. L. (2018). How
subjective and non-physical parameters affect occupants’ environmental comfort perception.
Energy and Buildings, 178, 107–129. https://doi.org/10.1016/J.ENBUILD.2018.08.020

Chen, C.-P., Hwang, R.-L., Chang, S.-Y ., & Lu, Y .-T. (2011). Effects of temperature steps on
human skin physiology and thermal sensation response. Building and Environment, 46(11),
2387–2397. https://doi.org/10.1016/j.buildenv.2011.05.021

Chen, J., Xue, J., Wang, Y ., Huang, L., Baker, T., & Zhou, Z. (2023). Privacy-Preserving and
Traceable Federated Learning for data sharing in industrial IoT applications. Expert Systems
with Applications, 213, 119036. https://doi.org/https://doi.org/10.1016/j.eswa.2022.119036

Choi, Y ., Kim, M., & Chun, C. (2015). Measurement of occupants’ stress based on
electroencephalograms (EEG) in twelve combined environments. Building and Environment,
88, 65–72. https://doi.org/10.1016/J.BUILDENV .2014.10.003

Cowan, J. (2014). Building Acoustics. In T. D. Rossing (Ed.), Springer Handbook of Acoustics (pp.
403–442). Springer New York. https://doi.org/10.1007/978-1-4939-0755-7_11

Dalton, N. S., Schnädelbach, H., Wiberg, M., & Varoudis, T. (2016). Architecture and interaction.

D’Ambrosio Alfano, F. R., Olesen, B. W., Palella, B. I., & Riccio, G. (2014). Thermal comfort:
Design and assessment for energy saving. Energy and Buildings, 81, 326–336.
https://doi.org/10.1016/J.ENBUILD.2014.06.033

Fernández, J. D., Menci, S. P., Lee, C. M., Rieger, A., & Fridgen, G. (2022). Privacy-preserving
93
federated learning for residential short-term load forecasting. Applied Energy, 326, 119915.
https://doi.org/https://doi.org/10.1016/j.apenergy.2022.119915

Flouris, A. D. (2011). Functional architecture of behavioural thermoregulation. European Journal
of Applied Physiology, 111(1), 1–8. https://doi.org/10.1007/s00421-010-1602-8

Hu, M., Simon, M., Fix, S., Vivino, A. A., & Bernat, E. (2021). Exploring a sustainable building’s
impact on occupant mental health and cognitive function in a virtual environment. Scientific
Reports, 11(1), 5644. https://doi.org/10.1038/s41598-021-85210-9

Huang, L., Zhu, Y ., Ouyang, Q., & Cao, B. (2012). A study on the effects of thermal, luminous,
and acoustic environments on indoor environmental comfort in offices. Building and
Environment, 49(1), 304–309. https://doi.org/10.1016/J.BUILDENV .2011.07.022

Jiang, D. (2020). The construction of smart city information system based on the Internet of Things
and cloud computing. Computer Communications, 150, 158–166.
https://doi.org/https://doi.org/10.1016/j.comcom.2019.10.035

Jobin, A., Ienca, M., & Vayena, E. (2019). The global landscape of AI ethics guidelines. Nature
Machine Intelligence, 1(9), 389–399. https://doi.org/10.1038/s42256-019-0088-2

Kahneman, D., & Deaton, A. (2010). High income improves evaluation of life but not emotional
well-being. Proceedings of the National Academy of Sciences, 107(38), 16489–16493.
https://doi.org/10.1073/pnas.1011492107

Kaplan, S. (1995). The restorative benefits of nature: Toward an integrative framework. Journal
of Environmental Psychology, 15(3), 169–182. https://doi.org/10.1016/0272-4944(95)90001-
2

Kim, H., Hong, T., Kim, J., & Yeom, S. (2020). A psychophysiological effect of indoor thermal
condition on college students’ learning performance through EEG measurement. Building and
Environment, 184, 107223. https://doi.org/10.1016/J.BUILDENV .2020.107223

Kim, M.-K., Kim, M., Oh, E., & Kim, S.-P. (2013). A Review on the Computational Methods for
Emotional State Estimation from the Human EEG. Computational and Mathematical
Methods in Medicine, 2013, 573734. https://doi.org/10.1155/2013/573734

Li, H., Li, C., Wang, J., Yang, A., Ma, Z., Zhang, Z., & Hua, D. (2023). Review on security of
federated learning and its application in healthcare. Future Generation Computer Systems.
https://doi.org/https://doi.org/10.1016/j.future.2023.02.021

94
Li, J., Jin, Y ., Lu, S., Wu, W., & Wang, P. (2020). Building environment information and human
perceptual feedback collected through a combined virtual reality (VR) and
electroencephalogram (EEG) method. Energy and Buildings, 224, 110259.
https://doi.org/10.1016/j.enbuild.2020.110259

Lo, S. K., Lu, Q., Zhu, L., Paik, H.-Y ., Xu, X., & Wang, C. (2022). Architectural patterns for the
design of federated learning systems. Journal of Systems and Software, 191, 111357.
https://doi.org/https://doi.org/10.1016/j.jss.2022.111357

McMahan, B., Moore, E., Ramage, D., Hampson, S., & Arcas, B. A. y. (2017). Communication-
Efficient Learning of Deep Networks from Decentralized Data. In A. Singh & J. Zhu (Eds.),
Proceedings of the 20th International Conference on Artificial Intelligence and Statistics (Vol.
54, pp. 1273–1282). PMLR. https://proceedings.mlr.press/v54/mcmahan17a.html

Melikov, A. K. (2004). Personalized ventilation. Indoor Air, 14(s7), 157–167.
https://doi.org/10.1111/j.1600-0668.2004.00284.x

Noda, N., Hetnarski, R. B., & Tanigawa, Y . (2018). Thermal Stresses. Routledge.
https://doi.org/10.1201/9780203735831

Pandya, S., Srivastava, G., Jhaveri, R., Babu, M. R., Bhattacharya, S., Maddikunta, P. K. R.,
Mastorakis, S., Piran, Md. J., & Gadekallu, T. R. (2023). Federated learning for smart cities:
A comprehensive survey. Sustainable Energy Technologies and Assessments, 55, 102987.
https://doi.org/https://doi.org/10.1016/j.seta.2022.102987

Parsons, K. (2007). Human Thermal Environments. CRC Press.
https://doi.org/10.1201/9781420025248

Perez-Pozuelo, I., Spathis, D., Clifton, E. A. D., & Mascolo, C. (2021). Wearables, smartphones,
and artificial intelligence for digital phenotyping and health. Digital Health, 33–54.
https://doi.org/10.1016/B978-0-12-820077-3.00003-1

Ramu, S. P., Boopalan, P., Pham, Q.-V ., Maddikunta, P. K. R., Huynh-The, T., Alazab, M., Nguyen,
T. T., & Gadekallu, T. R. (2022). Federated learning enabled digital twins for smart cities:
Concepts, recent advances, and future directions. Sustainable Cities and Society, 79, 103663.
https://doi.org/https://doi.org/10.1016/j.scs.2021.103663

Rolfö, L. V . (2018). Relocation to an activity-based flexible office – Design processes and
outcomes. Applied Ergonomics, 73, 141–150.
https://doi.org/10.1016/J.APERGO.2018.05.017

95
Shan, X., Yang, E.-H., Zhou, J., & Chang, V . W.-C. (2018). Human-building interaction under
various indoor temperatures through neural-signal electroencephalogram (EEG) methods.
Building and Environment, 129, 46–53. https://doi.org/10.1016/j.buildenv.2017.12.004

Tang, L., Xie, H., Wang, X., & Bie, Z. (2023). Privacy-preserving knowledge sharing for few-shot
building energy prediction: A federated learning approach. Applied Energy, 337, 120860.
https://doi.org/https://doi.org/10.1016/j.apenergy.2023.120860

Thakor, N. v, & Tong, S. (2004). Advances in Quantitative Electroencephalogram Analysis
Methods. Annual Review of Biomedical Engineering, 6(1), 453–495.
https://doi.org/10.1146/annurev.bioeng.5.040202.121601

Wang, X., Li, D., Menassa, C. C., & Kamat, V . R. (2019). Investigating the effect of indoor thermal
environment on occupants’ mental workload and task performance using
electroencephalogram. Building and Environment, 158, 120–132.
https://doi.org/10.1016/J.BUILDENV.2019.05.012

Wohlers, C., & Hertel, G. (2017). Choosing where to work at work – towards a theoretical model
of benefits and risks of activity-based flexible offices. Ergonomics, 60(4), 467–486.
https://doi.org/10.1080/00140139.2016.1188220

Wyon, D. P. (2004). The effects of indoor air quality on performance and productivity. Indoor Air,
14 Suppl 7, 92–101. https://doi.org/10.1111/j.1600-0668.2004.00278.x

Yang, A., Ma, Z., Zhang, C., Han, Y ., Hu, Z., Zhang, W., Huang, X., & Wu, Y . (2022). Review on
application progress of federated learning model and security hazard protection. Digital
Communications and Networks. https://doi.org/https://doi.org/10.1016/j.dcan.2022.11.006

Yao, Y ., Lian, Z., Liu, W., & Shen, Q. (2008). Experimental study on physiological responses and
thermal comfort under various ambient temperatures. Physiology & Behavior, 93(1–2), 310–
321. https://doi.org/10.1016/J.PHYSBEH.2007.09.012

Zhang, L., Wen, J., Li, Y ., Chen, J., Ye, Y ., Fu, Y ., & Livingood, W. (2021). A review of machine
learning in building load prediction. Applied Energy, 285, 116452.
https://doi.org/https://doi.org/10.1016/j.apenergy.2021.116452

Asset Metadata

Creator Yin, Xiaoyu (author)

Core Title Dynamic workplace platform: exploration of the feasibility of human electroencephalogram (EEG) to predict the user’s indoor environmental satisfaction

Contributor Electronically uploaded by the author (provenance)

School School of Architecture

Degree Master of Building Science

Degree Program Building Science

Degree Conferral Date 2023-05

Publication Date 04/27/2025

Defense Date 03/29/2023

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

Tag brain signals,human,indoor environmental quality,OAI-PMH Harvest,prediction model

Format theses (aat)

Language English

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

Creator Email 695446607@qq.com,xiaoyuyi@usc.edu

Unique identifier UC113089400

Identifier etd-YinXiaoyu-11733.pdf (filename)

Legacy Identifier etd-YinXiaoyu-11733

Document Type Thesis

Format theses (aat)

Rights Yin, Xiaoyu

Internet Media Type application/pdf

Type texts

Source 20230501-usctheses-batch-1033 (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.

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 uscdl@usc.edu

Abstract (if available)

Abstract As society becomes increasingly aware of the importance of indoor environmental quality (IEQ), the need for more comprehensive assessments of IEQ is growing. With the advent of new types of offices, such as dynamic office platforms, there is a pressing need to assess the impact of IEQ on human physiology. This study aims to investigate whether differences in human physiological responses are caused by variations in IEQ in different areas of the same office. Data collection was conducted at an innovative engineering firm in Southern California that already started adopting a dynamic work platform. This research collected brain signals from employees working in different areas and analyze the relationship between their EEG signals and various indoor environmental parameters, as well as their responses to questionnaires about their perceptions of the indoor environment. The results showed the influence of EEG signals and the indoor environment that can be visualized. The results of analysis could be used to develop a prediction model that can be used to visualize the influence of EEG signals on indoor environmental quality. This model will be applied to predict future indoor environmental parameters based on existing EEG data, allowing people to effectively improve the indoor environment in the future. The study found that EEG signals can be used to predict indoor temperature, but further research is needed to investigate the relationship between other physiological responses and indoor environmental parameters. This will lead to the development of a more comprehensive model for predicting and improving indoor environmental quality.