Base Paper
Base Paper
iew
Swaymprabha Alias Megha Mane1, Megha Patil2, Arundhati Shinde3
1
Department od Electronics and Communication Engineering, Bharati Vidyapeeth (Deemed to be University) College of Engineering, Pune, India
2
School of Mechatronics Engineering, Symbiosis Skills and Professional University, Pune, India
ABSTRACT: This article presents an EEG dataset collected using the EMOTIV Insight EEG 5-Channel
v
Sensor kit during five different types of stimulation: complex mathematical problem solving, horror video
watching, trier mental challenge test and stroop color word test, listening to relaxing music. The dataset
consists of EEG recordings from 22 subjects for mathematical problem solving, 22 subjects for horror video
re
stimulation, 24 subjects for trier mental challenge test, 24 subjects for the Stroop color word test, 20 subjects
for relaxed state recordings. The data was collected in order to investigate the neural correlates of stress and
to develop models for stress detection based on EEG data. In addition, StressNet a stress detection model was
developed and evaluated for deep learning applications. The method consisted of two parts: Azimuthal
Projection and CNN (Convolutional Neural Network) & LSTM (Long short-term memory) StressNet model.
er
The Azimuthal Projection technique was used to preprocess the raw EEG signals, and the CNN & LSTM
model was used to classify the signals into stressed and non-stressed categories. The performance of the stress
detection method was evaluated using various metrics, including accuracy, sensitivity, specificity, and F1
score. The dataset and stress detection method presented in this article can be used for various applications,
pe
including stress management, healthcare and workplace safety. The dataset provides a valuable resource for
researchers and developers working on stress detection using EEG data, while the stress detection method
provides a useful tool for evaluating the effectiveness of different stress detection models. Overall, this article
contributes to the growing body of research on stress detection and management using EEG data, and provides
a useful resource for researchers and practitioners working in this field.
ot
Keywords: Azimuthal Projection, EEG dataset, EMOTIV EEG sensor kit, Convolutional Neural Network
(CNN), Stress detection.
tn
stress is crucial for maintaining overall health and reducing Better Decision Making: Stress can impair decision making
the risk of developing stress-related illnesses. With the by reducing the ability to focus and think clearly. By
advancements in technology, stress detection has become detecting stress early on, individuals can take steps to reduce
easier and more accessible, and it is now possible to detect stress levels and make better decisions.
and monitor stress levels in real-time. Enhanced Quality of Life: Chronic stress can lead to a range
ep
Early Detection of Chronic Stress: Chronic stress can have of physical and mental health problems, including
detrimental effects on physical and mental health if left depression, anxiety, and cardiovascular disease. By detecting
unaddressed. By detecting stress early on, individuals can stress early on and managing it effectively, individuals can
take proactive steps to manage their stress levels and prevent improve their overall quality of life and reduce the risk of
chronic stress from developing [2]. developing health problems associated with chronic stress.
Stress is a ubiquitous part of modern life, and it can have a
Pr
This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4724661
electroencephalography (EEG), which measures the associated with reduced stress, or to test the effectiveness of
ed
electrical activity of the brain and can be used to detect stress different stress reduction techniques (such as mindfulness
[3]. meditation) on brain activity and stress levels.
In this article, a dataset of EEG recordings that were
collected with the help of EMOTIV insight portable EEG II. RELATED WORK
sensor kit. The dataset consists of EEG recordings from The field of electroencephalogram (EEG) recording,
iew
individuals who were subjected to a variety of stress- particularly employing the 10-20 based EEG recording
inducing tasks. The dataset also includes recordings from a systems, has witnessed remarkable advancements, leading to
control group who were not subjected to any stress-inducing enhanced capabilities in identifying various brain activities
tasks. The purpose of this dataset is to provide researchers and diseases. These systems have proven instrumental in
and practitioners with a valuable resource for studying stress revolutionizing the landscape of medical assistive
detection using EEG. The EMOTIV insight portable EEG technologies by offering non-contact methods for capturing
sensor kit [4] is an affordable and accessible tool that can be EEG signals and subsequently processing them to identify
used in a variety of settings, including research labs and characteristic features associated with specific mental states
v
clinical settings. The dataset includes a range of stress- or diseases. Notably, this includes conditions such as
inducing tasks, which can help researchers better understand alertness, dizziness, sleep states, stress, and anxiety.
re
how the brain responds to stress in different contexts. The allure of non-contact EEG signal capturing lies in its
Value of Data are, the Stress EEG dataset recorded using ability to provide a nuanced understanding of brain function
EMOTIV INSIGHT headset can potentially be used by without invasive measures. This is especially crucial in
researchers, scientists, and students who are interested in medical applications, where minimizing patient discomfort
studying brain activity related to stress. This dataset may also and ensuring convenience are paramount. The processing of
be useful for professionals working in the fields of EEG signals typically involves a multi-stage approach,
er
neuroscience, psychology, and psychiatry who are interested
in understanding the effects of stress on brain activity.
Proposed dataset consists of 5 sub-datasets which are created
by the research point of view with total 112 EDF files which
encompassing preprocessing, feature extraction, and
classification.
In the preprocessing stage, raw EEG data undergoes various
techniques to enhance its quality and extract relevant
pe
can be further increased with another different data information. This may involve filtering out noise, correcting
augmentation processes. It is important to note that the use artifacts, and normalizing signals to facilitate more accurate
of this dataset may require specific knowledge and expertise analysis in subsequent stages. Feature extraction follows,
in the field of EEG data analysis. wherein distinctive attributes of the EEG signal related to
Additionally, it is important to ensure that any research specific mental states or diseases are identified. This phase
conducted using this dataset follows ethical guidelines and is crucial for distilling relevant information from the
regulations for the use of human subjects in research. The complex EEG data.
ot
availability of this dataset can help accelerate research in the Classification, the final stage, involves the use of
field of stress detection and contribute to the development of sophisticated algorithms to categorize EEG signals into
more effective stress management tools. predefined classes, such as different brain states or the
Objective of this research work are, investigating the presence of specific disorders. Nonlinear classifiers,
tn
relationship between brain activity and stress: Researchers including but not limited to K nearest neighbor, support
may use the dataset to explore how changes in brain activity vector machines, and other machine learning-based
relate to the experience of stress, and to identify patterns of algorithms, have demonstrated notable success in this
brain activity that are associated with different types or levels domain. Additionally, the rise of deep neural network-based
of stress. methods has further expanded the horizons of EEG signal
Developing methods for measuring stress using EEG: EEG processing, with these models exhibiting commendable
rin
data can be used to measure brain activity in real-time, performance in accurately classifying complex patterns
making it a potentially useful tool for monitoring stress within the data.
levels in individuals. Researchers may work with the dataset The focus on brain state or activity identification using EEG
to develop new methods for using EEG to measure stress, or signals aligns with broader efforts in medical diagnosis,
to improve existing methods. where leveraging signal processing and neural network
ep
Exploring the effects of stress on cognitive function: Stress platforms has become pivotal. The integration of these
can have a significant impact on cognitive function, and technologies into healthcare systems holds immense
researchers may use the dataset to explore how stress affects potential for enhancing diagnostic accuracy, enabling earlier
different aspects of cognitive function (such as attention, detection of disorders, and facilitating personalized
memory, or decision-making). treatment approaches. The proposed work, situated within
this context, contributes to the evolving landscape of EEG-
Pr
This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4724661
and management of neurological conditions. As technology based methods offer a more dynamic and adaptive approach,
ed
continues to evolve, the synergy between EEG recording capable of capturing intricate patterns and nuances within
systems and advanced processing algorithms promises a EEG signals that may elude rule-based systems.
future where non-invasive monitoring of brain activity The advantages of DL-based EEG signal processing extend
becomes an indispensable tool in the realm of healthcare. beyond improved accuracy and efficiency. The adaptability
In contemporary healthcare, medical imaging-based EEG of DL models to evolving datasets and their capacity to
iew
signal processing has emerged as a transformative force, handle diverse and complex information make them well-
demonstrating substantial impact in the market. This suited for the inherently dynamic nature of EEG signals.
influence is attributable to the revolutionary strides made Furthermore, the robustness of DL-based methods in the face
possible by virtualization technology, wherein EEG signals of challenges such as artifacts and non-stationary signals
are manipulated without the need for the actual performance positions them as formidable tools in the quest for more
of associated tasks, as highlighted in [4]. This paradigm shift reliable and accurate medical diagnostics.
in EEG signal processing has opened up new vistas, As medical imaging-based EEG signal processing continues
promising enhanced accuracy and efficiency in the to advance, the shift towards DL methodologies marks a
v
identification of characteristic features related to specific significant stride towards unlocking the full potential of this
tasks or mental states. technology. The confluence of virtualization technology,
re
One of the primary challenges in dealing with EEG signals multi-channel recording paradigms, and deep learning
lies in the dimensionality of the data. The temporal aspects heralds a new era in healthcare, where the non-invasive
of recordings, the presence of artifacts, non-stationary monitoring and processing of EEG signals promise not only
signals, and the intricate correlations between channels in enhanced diagnostic capabilities but also a deeper
multi-channel recording setups contribute to the complexity understanding of the intricacies of neural function. In the
of the data [6]. These challenges are particularly pronounced pursuit of overcoming the limitations of traditional neural
efficacy has been hampered by inherent limitations, notably floating-point matrix multiplications [8-10]. The
the dependence on handcrafted rules for feature extraction computational demands associated with processing large
[8]. As a result, there has been a persistent need for more datasets have been effectively addressed through the
advanced and adaptive methodologies to overcome the integration of graphical processing units (GPUs),
tn
limitations of traditional neural networks. overcoming the challenges posed by the complexity of deep
This need has given rise to the ascendancy of deep learning learning methods and paving the way for their successful
(DL) based methods in the realm of EEG signal processing. application in medical diagnostics [14].
DL methods, characterized by their ability to automatically The application of deep neural networks (DNNs) to process
learn hierarchical representations from data, have exhibited machine-interfaced signals has garnered significant attention
a notable leap in accuracy and robustness compared to their among researchers [15]. These applications extend to a
rin
traditional counterparts. The intrinsic capability of DL diverse range of fields, including the control of wheelchairs
models to discern complex patterns within EEG signals or robots [16], assistance for paralyzed patients in walking
without relying on explicitly defined rules has proven to be [17], and the development of upper limb exoskeletons [18],
a game-changer. Moreover, the efficiency of DL-based among others [7]. However, the unique characteristics of
methods is underscored by their reduced processing time and motor imagery (MI) signals, characterized by small
ep
a diminished requirement for meticulous tuning of amplitudes and a mixture of artifacts, present challenges that
parameters, overcoming one of the key limitations of can degrade the performance of the systems [19, 20].
traditional neural networks [9]. In addressing these challenges, Li et al. [21] explored the use
In the pursuit of refining EEG signal processing, it is of CNNs for signal classification by extracting temporal
essential to underscore the shortcomings of traditional features from EEG signals. The CNN architecture was adept
Pr
approaches. The reliance on handcrafted rules, coupled with at capturing high-level features critical for discerning
the time-consuming nature of processing and the patterns indicative of specific mental states or tasks. Building
susceptibility to information loss, has impeded the full on this, Alazrai et al. [4] proposed a method of obtaining 2D
potential of these methodologies [10]. On the contrary, DL- images from EEG signals, capitalizing on the energy
This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4724661
distribution over time and frequency domains using Fast approach involved multivariate empirical mode
ed
Fourier Transform (FFT) techniques [23]. This approach not decomposition, enabling the extraction of intrinsic mode
only facilitated efficient data representation but also function signals from recorded multichannel EEG data. The
enhanced the interpretability of the neural network. notable achievement of a 93% f1 score demonstrates the
The exploration of specific CNN architectures for EEG effectiveness of this method in capturing distinctive features
signal processing has been a focal point of research. Xu et al. associated with schizophrenia conditions in EEG signals.
iew
[8] employed the VGG-16 model, a deep CNN architecture, Shifting focus to cardiac health, Murugappan et al. [27]
for their studies. To address the challenge of data scarcity explored the application of signal processing for heart
and the need for substantial labeled datasets, transfer disease detection using ECG signals. Analyzing the PQRST
learning was incorporated into the training process. This features of the ECG, with specific attention to the R to T end
technique leverages pre-trained models on large datasets and features, the method aimed to identify sudden cardiac arrests
fine-tunes them for specific tasks, thereby optimizing the from recorded ECG signals. The utilization of K-nearest
model's performance with limited labeled EEG data. neighbor and support vector machine classifiers underscored
Zhao et al. [23] took a multi-branch CNN approach for the the versatility of these machine learning techniques in
v
classification of MI signals. In their innovative network medical diagnostics. This approach not only showcased the
architecture, EEG data is fed into three separate CNNs in adaptability of signal processing methodologies across
re
parallel, each processing the input independently. The final different physiological domains but also highlighted the
classification result is determined by aggregating the importance of tailored algorithms for specific medical
outcomes obtained from the three CNN branches. This multi- applications.
branch strategy enhances the robustness and generalization The incorporation of machine learning classifiers, such as
capabilities of the model by capturing diverse aspects of the SVM-RBF and K-nearest neighbor, in the context of EEG
input data. and ECG signal processing reflects the interdisciplinary
er
Despite the significant strides made in the application of
deep learning to EEG signal processing, challenges persist.
The intricate nature of neural signals demands continuous
refinement and adaptation of methodologies. The small
nature of contemporary medical informatics. These
classifiers leverage the power of algorithms trained on
diverse datasets to generalize patterns and make accurate
predictions, thereby enhancing diagnostic accuracy.
pe
amplitudes and susceptibility to artifacts in MI signals The unique advantage of EEG signals lies in their ability to
necessitate ongoing efforts to improve the robustness of capture the intricacies of brain function, making them
models. Additionally, the integration of advanced techniques invaluable for diagnosing neurological conditions. By
such as CWT, FFT, and innovative network architectures extending the application of EEG signals beyond motor
reflects the interdisciplinary nature of this research, where imagery to conditions like schizophrenia, researchers
insights from signal processing and machine learning contribute to a more comprehensive understanding of the
converge to push the boundaries of what is achievable in diagnostic potential inherent in these signals. Similarly, in
ot
EEG-based diagnostics. the domain of cardiac health, the analysis of ECG signals has
In the dynamic landscape of EEG signal processing, Tabar become a critical component in the early detection of heart
and Halici [23] proposed an innovative approach by diseases and abnormalities.
combining Convolutional Neural Network (CNN) and The combination of advanced neural network architectures,
tn
stacked autoencoders (SAE). The utilization of the short- such as CNN and SAE, with sophisticated signal processing
time Fourier transform (STFT) to construct 2D images for techniques like STFT, multivariate empirical mode
network training highlighted the versatility of signal decomposition, and machine learning classifiers showcases
processing techniques. In their framework, CNNs were the evolution of EEG and ECG signal processing. These
employed for extracting features within motor imagery (MI) methodologies not only enhance the accuracy of diagnostics
signals, and these features were subsequently classified but also pave the way for personalized medicine by tailoring
rin
using SAE. This integrated approach showcases the synergy approaches to the unique characteristics of individual
between different neural network architectures and signal patients.
processing methodologies, emphasizing the importance of a As the field of medical informatics continues to advance, the
holistic strategy in harnessing the full potential of EEG integration of diverse signal processing techniques and
signals. machine learning methodologies promises to revolutionize
ep
Beyond the realm of motor imagery, EEG signals have found diagnostics. The ability to harness the wealth of information
application in probing complex neurological conditions such embedded in EEG and ECG signals opens new horizons for
as schizophrenia. Schizophrenia, characterized by extreme understanding and treating complex neurological and
levels of hallucination and unreal behaviors, presents a cardiovascular conditions. In essence, the convergence of
unique challenge for diagnosis and analysis. Palani et al. [26] signal processing and machine learning in the analysis of
Pr
delved into the realm of schizophrenia detection by utilizing physiological signals represents a powerful paradigm shift,
EEG signals. They employed a radial basis function support offering unprecedented insights into the intricate workings
vector machine (SVM-RBF) classifier to discern patterns of the human body and mind.
indicative of schizophrenia conditions. The processing
This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4724661
Figure 1: Mathematical Problem Solving to identify register
ed
value by using color code technique
II. MATERIALS AND METHODS
iew
EEG signals have been described as well.
A. Dataset Description
1) DATASET 1 MATHEMATICAL PROBLEM SOLVING:
Mathematical problem-solving can give rise to stress in
several ways [28]. Here are some of the ways in which
mathematical problem-solving can lead to stress:
Cognitive Load: Mathematical problem-solving often
v
requires the use of working memory to hold information in
mind while solving the problem. This can lead to cognitive (a)
re
load, which is the amount of mental effort required to
complete a task. When the cognitive load is high, it can cause
stress and mental fatigue.
Time Pressure: Many math problems are timed, which can
cause stress if subject feel like they don't have enough time
to solve the problem. Time pressure can also make subjects
feel rushed, which can cause them to make mistakes or
overlook important details.
er
Fear of Failure: Mathematics can be challenging, and it's not
uncommon to experience fear of failure when solving math
pe
problems. This fear can lead to stress and anxiety, which can
make it difficult to concentrate and solve the problem.
Lack of Confidence: When subjects lack confidence in their
math abilities, it can be stressful to try to solve math
problems. Subject feels like they don't know where to start.
This lack of confidence can make it difficult to focus and can (b)
increase stress levels. Figure 2: Subjects solving mathematical problems on
ot
Perfectionism: Perfectionism is the tendency to set high integration and different operations on signals
standards for yourself and to strive for perfection in
everything they do. When this mindset applies to solving Complex mathematical problem-solving shown in figure 2
math problems, it can lead to stress and anxiety. Subject tried can give rise to stress in several ways, including cognitive
tn
to solve the problem perfectly, which can cause them to load, time pressure, fear of failure, lack of confidence, and
become hyper-focused on details. perfectionism. However, with practice and proper strategies,
The example of Stress inducing mathematical problem it is possible to reduce stress levels and improve
solving considered while recording EEG signals is Resistor mathematical problem-solving abilities.
Color Code as shown in figure 1. and subject solving
rin
complex mathematical problems such as integration and Sr. Participants Gender Age Duration of
different operations on signals shown in figure 2 No. Years Recording
1 Subject 1 Male 23 3 Min 48 Sec
2 Subject 2 Male 21 5 Min 36 Sec
3 Subject 3 Male 21 4 Min 09 Sec
ep
This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4724661
13 Subject 13 Male 23 19 Min 14 Sec dataset. Horror videos, in particular, are an effective stress-
ed
14 Subject 14 Male 21 8 Min 06 Sec inducing stimulus that can be standardized and controlled,
15 Subject 15 Male 21 10 Min 22 Sec and used in a variety of contexts. The use of horror videos
16 Subject 16 Male 19 5 Min 05 Sec for EEG-based stress detection dataset preparation can have
17 Subject 17 Male 21 10 Min 19 Sec significant practical applications, such as in training
18 Subject 18 Male 21 4 Min 54 Sec individuals who are exposed to stressful or dangerous
iew
19 Subject 19 Male 21 9 Min 10 Sec situations. Overall, EEG-based stress detection using
20 Subject 20 Male 21 4 Min 39 Sec simulated video-based stress induction has significant
21 Subject 21 Male 20 8 Min 58 Sec potential for improving our understanding of stress and its
impact on human health and well-being.
22 Subject 22 Female 23 7 Min 17 Sec
v
2) DATASET 2 STIMULATED STRESS USING HORROR
VIDEOS:
re
Stimulated video-based stress induction is a commonly used
method for inducing stress in laboratory settings [29]. This
method involves exposing participants to videos that are
designed to induce a stress response. In recent years,
researchers have started using EEG-based stress detection
methods to detect and monitor stress levels in response to (a)
er
these videos. This approach has significant potential for
creating a valuable EEG-based stress detection dataset.
One type of video that has been found to be particularly
effective in inducing stress is horror videos as shown in
figure 3. Horror videos are designed to evoke fear and
pe
anxiety in viewers, and they can be a powerful tool for
inducing stress in laboratory settings. In this dataset, horror
videos are standardized and controlled, ensuring that all
participants are exposed to the same level of stress. This
improved the reliability and validity of the dataset
Secondly, horror videos induced stress in a safe and
ot
dataset for stress detection. Furthermore, horror videos can 3 Subject 3 Male 34 01 Min 59 Sec
be used in a variety of contexts, such as in response to 4 Subject 4 Male 16 01 Min 08 Sec
different themes or situations, allowing for a more 5 Subject 5 Male 22 02 Min 51 Sec
comprehensive dataset. In addition to creating a valuable 6 Subject 6 male 68 00 Min 50 Sec
dataset for stress detection, horror videos can also have 7 Subject 7 Female 65 01 Min 16 Sec
ep
practical applications in the real world. For example, horror 8 Subject 8 Female 35 02 Min 51 Sec
videos could be used to train individuals who are exposed to 9 Subject 9 Male 22 01 Min 39 Sec
stressful or dangerous situations, such as soldiers or 10 Subject 10 Male 19 02 Min 57 Sec
emergency responders. By exposing individuals to stress in 11 Subject 11 Male 21 02 Min 43 Sec
a controlled and safe environment, they can develop coping 12 Subject 12 Male 16 01 Min 12 Sec
mechanisms and improve their ability to perform under
Pr
This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4724661
17 Subject 17 Male 21 01 Min 12 Sec how the Stroop Effect is affected by factors like age,
ed
18 Subject 18 Male 39 01 Min 14 Sec language, and working memory capacity. They have also
19 Subject 19 Male 37 01 Min 09 Sec used the Stroop task to study cognitive processes like
20 Subject 20 Male 20 01 Min 19 Sec attentional control, response inhibition, and cognitive
21 Subject 21 Female 23 01 Min 53 Sec flexibility.
22 Subject 22 Female 31 03 Min 29 Sec The Stroop Effect has also been used in clinical settings
iew
to study disorders like attention deficit hyperactivity disorder
Table 2: Summery of Stimulated Stress using Horror Videos (ADHD) and schizophrenia. In ADHD, for example, people
may have difficulty inhibiting automatic responses and
focusing attention, which can make the Stroop task more
3) DATASET 3 STROOP COLOUR WORD TEST:
challenging. In schizophrenia, people may have difficulty
The Stroop Effect is a cognitive phenomenon that occurs
processing conflicting information, which can also make the
when there is a mismatch between the color of a word and
Stroop task more difficult.
the word itself. Specifically, it refers to the delay in reaction
Overall, the Stroop Effect is a powerful tool for studying
v
time that occurs when people are asked to name the color of
cognitive processes and has led to many important insights
the ink that a word is printed in, rather than reading the word
in the field of psychology. It has also been applied in a
itself.
re
variety of settings, from basic research to clinical practice,
The Stroop Effect was first described by John Ridley
and has helped researchers better understand the
Stroop in 1935 [30]. In his original experiment, participants
complexities of the human mind.
were presented with a list of color names printed in black ink,
The Stroop Effect refers to the delay in reaction time that
and they were asked to name the ink color as quickly as
occurs when the color of a word and the word itself are
possible. Stroop found that people took longer to name the
mismatched. This can result in difficulty in reading the word
er
ink color when the color name was printed in a color that did
not match the name (e.g., the word "red" printed in blue ink).
The Stroop Effect has been replicated in many studies
since then, and it has become an important tool for studying
or naming the color of the ink, and can lead to increased
cognitive load and stress.
For example, if the word "black" is written in white ink,
it can take longer to read the word correctly because your
cognitive processes like attention and interference. In the
pe
brain is processing conflicting information. Your brain has
classic version of the Stroop task, participants are presented
to process the meaning of the word "black" and then override
with a list of words printed in different colors and are asked
that information with the color of the ink, which is white.
to name the ink color while ignoring the word itself. The
This can cause cognitive conflict and increase stress levels.
words can be color names (e.g., "red," "green," "blue"), or
The Stroop Effect is often used as a psychological test
they can be neutral words (e.g., "chair," "table," "house") as
to measure attentional control and cognitive flexibility. It can
shown in figure 4.
also be used to study the effects of stress on cognitive
ot
mismatch, the automatic process of reading the word 2 Subject 2 Male 19 2 Min 47 Sec
interferes with the effortful process of naming the ink color, 3 Subject 3 Male 39 2 Min 00 Sec
leading to a delay in reaction time. 4 Subject 4 Male 21 3 Min 26 Sec
5 Subject 5 Male 40 3 Min 40 Sec
6 Subject 6 Male 21 4 Min 05 Sec
rin
The Stroop Effect has been studied extensively in 18 Subject 18 Male 39 2 Min 05 Sec
cognitive psychology, and it has been used to investigate a 19 Subject 19 Male 37 3 Min 41 Sec
variety of topics [8]. For example, researchers have studied 20 Subject 20 Male 20 2 Min 12 Sec
This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4724661
21 Subject 21 Female 23 3 Min 30Sec
ed
22 Subject 22 Male 22 1 Min 40 sec
23 Subject 23 Male 23 2 Min 12 Sec
24 Subject 24 Female 36 2 min 02 Sec
iew
4) DATASET 4 TRIED MENTAL CHALLENGE TEST:
Mental challenge tests can be used to detect stress by
measuring changes in physiological and cognitive responses
to mental tasks [32]. When people experience stress, their
bodies respond with physiological changes such as increased
heart rate, respiration rate, and blood pressure. These
v
changes can be measured using various physiological (b)
sensors such as electrocardiogram (ECG), Figure 5: Timed Mental Challenge Test
electroencephalogram (EEG), and skin conductance sensors. The results from these mental challenge tests can
re
In addition, stress can also affect cognitive performance, provide valuable information for stress detection and
such as memory and attention, which can be measured using management. For example, if someone shows consistently
various cognitive tests. high physiological and cognitive stress responses to mental
One example of a mental challenge test that can be used tasks, it may indicate a chronic stress condition that requires
for stress detection is the Stroop Test. As mentioned earlier, further evaluation and treatment. On the other hand, if
MUSIC:
laboratory procedure that involves public speaking and In recent years, the use of EEG recording has become
mental arithmetic tasks. During the test, participants are increasingly popular in both clinical and research settings.
asked to give a speech and perform mental arithmetic tasks EEG recording allows for the measurement of brain activity,
in front of an audience. This task is designed to induce stress, which can provide valuable insights into various mental
tn
and physiological and cognitive responses are measured states, including relaxation and meditation. One potential
before, during, and after the test. The example of arithmetic challenge with EEG recording is inducing a relaxed state in
time-based mental challenge test is shown in Figure 5. participants. Many individuals may find it difficult to relax
and stay still for the duration of the recording, which can
affect the quality of the data. Therefore, researchers have
rin
This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4724661
to calming and soothing music can induce a relaxed state for out unwanted noise and artifacts, and can achieve high
ed
EEG recording. accuracy rates even when the data contains these types of
The study also found that the choice of music was critical in artifacts.
achieving the desired level of relaxation. Slow, gentle music
with a soft melody was found to be most effective in C. Events/Artifacts are Important
inducing a relaxed state. This finding highlights the Muscular movements and eye opening/closing events
iew
importance of selecting appropriate music to achieve the can play an important role in stress detection when analyzing
desired level of relaxation and emphasizes the potential EEG signals [34]. While these events can introduce artifacts
benefits of using music to reduce stress and anxiety in into the data, they can also provide valuable information
clinical and research settings. about the individual's physiological response to stress.
Muscular movements, such as clenching of the jaw or
tightening of the muscles, can be associated with stress and
anxiety. These movements can be detected using
electromyography (EMG) sensors, which measure the
v
electrical activity of muscles. The frequency and intensity of
muscular movements can provide insight into the level of
re
stress experienced by the individual, as well as the timing
and duration of stress responses.
Similarly, eye opening and closing events can be
associated with stress and can also be measured using EEG.
The frequency of these events can provide information about
the individual's level of arousal and alertness, as well as their
Deep learning algorithms are designed to automatically between different types of brain activity associated with
learn the underlying patterns in data, and can effectively stress and relaxation. By taking these events into account
filter out unwanted noise and artifacts in EEG signals [34]. when analyzing EEG data, researchers can develop more
These algorithms are trained using large datasets of EEG accurate and effective methods for stress detection, and
tn
recordings, and can learn to distinguish between different ultimately improve our understanding of the effects of stress
types of signals, including those associated with eye opening on the human body and mind.
and closing events. EEG (electroencephalography) is a technique that
Furthermore, recent studies have shown that deep measures the electrical activity of the brain using electrodes
learning algorithms can achieve high accuracy rates for placed on the scalp. EEG is widely used in research to study
stress detection even when the data contains artifacts such as
rin
the use of deep learning methods for stress detection can dataset is carried out using the EMOTIV INSIGHT portable
provide an effective solution for processing EEG data even EEG sensor module [4] during five different tasks/stimuli:
when these artifacts are present. In fact, the use of deep stimulated horror videos, mathematical problem solving,
learning methods may even be preferable in certain Stroop Effect based recordings, mental challenge work and
situations, as it can provide a more automated and efficient relaxed state. Table 1 shows the EMOTIV insight EEG
approach to analyzing large datasets of EEG recordings [12].
Pr
This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4724661
EEG sensors 5 channels: AF3, AF4, T7, T8, Pz
ed
Subject EEG mental stress detection dataset
2 references: CMS/DRL references on
left mastoid process
Type of EEG Signal recording using 5 channel
New three prong gummy sensor: for
data EMOTIV Insight advanced brainwear for
better hair penetration on Pz
brain computer interface (5 channels:
Sensor material: Hydrophilic semi-
AF3, AF4, T7, T8, Pz and 2
iew
dry polymer
references: CMS/DRL references on left
mastoid process)
Connectivity Wireless: Bluetooth Low Energy –
Improved radio connectivity, updated How the Subject belonging combination of
antenna with Bluetooth 5 support. data were different age group where considered
acquired within range 12 years to 68 years. During
EEG signals Sampling rate: 128 samples per recording EMOTIV insight sensor kit was
second per channel
v
used by allocating specific task to subject
Resolution: 16 bits with 1 LSB = to five different recordings where obtained
0.1275μV in which first was relax state condition and
re
Frequency response: 0.5-43Hz, digital remaining four where stressful conditions.
notch filters at 50Hz and 60Hz The stressful conditions achieved by
Filtering: Built in digital 5th order Sinc allocating task of complex mathematical
filter problem solving, horror video stimulation,
Dynamic range (input stroop color word test, trier mental
referred): 8400 μV(pp) challenge test
Motion sensors
Coupling mode: AC coupled
https://data.mendeley.com/datasets/wnshb
vdxs2 [36]
Table 1: specifications of EEG signal data collection
This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4724661
IV. Result and Analysis
ed
As per our proposed previous work presented in Megha et al. The evaluation metrics of the combined dataset reveal
[35], a hybrid model with use of LSTM and CNN is used to remarkable results, with a specificity of 98.3%, indicating a
validate the dataset with prediction performance. high proportion of correctly identified negative instances.
Furthermore, the sensitivity, which measures the ability to
correctly detect positive instances, scored 97.2%, indicating
iew
a high success rate in identifying mental challenges. The F1
score, which balances precision and recall, reached an
outstanding value of 98, reflecting the harmonious balance
achieved by our model.
v
re
Figure 9: Performance of Different Datasets using
different performance parameters
er
With the utilization of LSTM (Long Short-Term Memory)
and Azimuthal projection-based CNN (Convolutional
pe
Neural Network) models for evaluating a self-recorded
dataset, our research findings indicate that the accuracy for
mental challenges surpasses that of horror videos,
mathematical problem solving, and the Stroop test. The
implementation of these advanced models resulted in an Figure 11: Confusion Matrix for Self Dataset using model
impressive overall accuracy of 97.2% on the combined trained on DASPS
dataset, showcasing enhanced performance. The
ot
This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4724661
challenges and carefully curating the dataset, we were able
ed
to achieve such promising results. Future research could
focus on expanding the dataset even further, incorporating
more variations of mental challenges and considering
additional evaluation metrics to gain a more comprehensive
understanding of model performance.
iew
V. Conclusion
In conclusion, the recorded dataset for stress detection using
the EMOTIV sensor kit presents a valuable resource for
researchers and practitioners in the field of affective
computing. The dataset comprises data collected from a
group of participants, who were exposed to different stress-
inducing stimuli, including mental arithmetic tasks,
v
emotional videos, and physical activities.
The EMOTIV sensor kit provides a non-invasive and
re
convenient means of measuring physiological responses
associated with stress. The recorded data includes raw EEG
signals, as well as features extracted from the EEG signals.
Figure 13: Confusion Matrix for Self Dataset using model The dataset can be used to develop and validate machine
trained on SEED dataset learning models for stress detection, as well as to investigate
the relationship between stress and physiological responses.
er The dataset may also be useful for developing personalized
stress management interventions and for evaluating the
effectiveness of such interventions.
However, it is important to note that the dataset has some
pe
limitations, including a relatively small sample size and the
lack of diversity in the participant population. Additionally,
the dataset was collected under controlled laboratory
conditions, which may not fully capture the complexity and
variability of real-world stress experiences.
Despite these limitations, the recorded dataset for stress
detection using the EMOTIV sensor kit represents a valuable
ot
trained on Combined dataset the Stroop test. The impressive performance on the
combined dataset, with an accuracy of 97.2%, specificity of
These findings demonstrate the power of LSTM and 98.3%, sensitivity of 97.2%, and F1 score of 98, underscores
Azimuthal projection-based CNN models in accurately the potential of these datasets for development of models in
evaluating a diverse range of challenges. The higher various applications in cognitive assessment, mental health
ep
accuracy in mental challenges suggests that the models diagnostics, and beyond.
possess a strong ability to identify and analyze cognitive and
psychological patterns. This could potentially lead to the VI. References
development of effective tools for mental health monitoring [1] H. Yaribeygi, Y. Panahi, H. Sahraei, T. P. Johnston,
and intervention. and A. Sahebkar, “The impact of stress on body
function: A review,” EXCLI J., vol. 16, p. 1057, Jul.
Pr
It is important to note that the success of our models relies 2017, doi: 10.17179/EXCLI2017-480.
heavily on the quality and diversity of the self-recorded [2] G. A. Light et al., “Electroencephalography (EEG)
dataset. By ensuring the inclusion of a wide range of mental and Event-Related Potentials (ERP’s) with Human
Participants,” Curr. Protoc. Neurosci., vol.
This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4724661
CHAPTER, no. SUPPL. 52, p. Unit, Jul. 2010, doi: (2018) 1–13,
ed
10.1002/0471142301.NS0625S52. https://doi.org/10.1016/j.cmpb.2018.04.005.
[3] “Insight Brainwear® 5 Channel Wireless EEG [15] B. Rim, N. Sung, S. Min, M. Hong, Deep learning
Headset - EMOTIV.” in physiological signal data: a survey, Sensors
https://www.emotiv.com/insight/ (accessed Apr. (Switzerland) 20 (4) (2020) 969,
01, 2023). https://doi.org/10.3390/ s20040969.
iew
[4] R. Alazrai, M. Abuhijleh, H. Alwanni, M.I. Daoud, [16] R. Ron-angevin, F. Velasco-´alvarez, ´A.
A deep learning framework for decoding motor Fern´andez-rodríguez, A. Díaz-estrella, M. J.
imagery tasks of the same hand using EEG signals, Blanca-mena, F.J. Vizcaíno-martín, Brain-
IEEE Access 7 (2019) 109612–109627, computer Interface application: auditory serial
https://doi.org/10.1109/access.2019.2934018. interface to control a two- class motor-imagery-
[5] H.K. Lee, Y. Choi, Application of continuous based wheelchair, J. Neuroeng. Rehabil. 14 (2017)
wavelet transform and convolutional neural 1–16, https://doi.org/10.1186/s12984-017-0261-y.
network in decoding motor imagery brain-computer [17] H. Yu, H. Lu, S. Wang, K. Xia, Y. Jiang, P. Qian,
v
interface, Entropy 21 (12) (2019) 1–12, A general common spatial patterns for EEG
https://doi.org/10.3390/e21121199. analysis with applications to vigilance detection,
re
[6] S. Sakhavi, C. Guan, S. Yan, Learning temporal IEEE Access 7 (2019) 111102–111114,
information for brain-computer interface using https://doi.org/10.1109/access.2019.2934519.
convolutional neural networks, IEEE Trans. neural [18] D.O. Souto, T.K.F. Cruz, K. Coutinho, A. Julio-
networks Learn. Syst. 29 (11) (2018) 5619–5629. Costa, P.L.B. Fontes, V.G. Haase, Effect of motor
[7] N. Padfield, J. Zabalza, H. Zhao, V. Masero, J. Ren, imagery combined with physical practice on upper
EEG-based brain-computer interfaces using motor- limb rehabilitation in children with hemiplegic
(Switzerland) 19 (6)
https://doi.org/10.3390/s19061423.
er
imagery: techniques and challenges, Sensors
(2019) 1–34,
[10] W. Qiao, X. Bi, Deep spatial-temporal neural effective computing framework for classification of
network for classification of EEGbased motor motor imagery EEG signals for brain – computer
imagery, ACM International Conference interface. Advances in Computational Intelligence
Proceeding Series (2019) 265–272, Techniques, Springer, Singapore, 2020, pp. 17–35.
tn
https://doi.org/10.1145/3349341.3349414. [21] Y. Li, X.R. Zhang, B. Zhang, M.Y. Lei, W.G. Cui,
[11] I. Majidov, T. Whangbo, Efficient classification of Y.Z. Guo, A channel-projection mixed-scale
motor imagery electroencephalography signals convolutional neural network for motor imagery
using deep learning methods, Sensors (Switzerland) EEG decoding, IEEE Trans. Neural Syst. Rehabil.
19 (7) (2019) 1–13, Eng. 27 (6) (2019) 1170–1180,https://doi.org/
https://doi.org/10.3390/s19071736. 10.1109/TNSRE.2019.2915621.
rin
[12] S.H. Khan, M. Hayat, F. Porikli, Regularization of [22] N. Lu, T. Li, X. Ren, H. Miao, A deep learning
Deep Neural Networks with Spectral Dropout, scheme for motor imagery classification based on
Neural Netw. 110 (February 2019) (2019) 82–90, restricted boltzmann machines, IEEE Trans. Neural
https://doi. org/10.1016/j.neunet.2018.09.009. Syst. Rehabil. Eng. 25 (6) (2017) 566–576,
[13] E. Nurvitadhi, et al., Can FPGAs beat GPUs in https://doi.org/10.1109/ TNSRE.2016.2601240.
ep
accelerating next-generation deep neural networks? [23] X. Zhao, H. Zhang, G. Zhu, F. You, S. Kuang, L.
FPGA 2017 - Proceedings of the 2017 Sun, A multi-branch 3D convolutional neural
ACM/SIGDA International Symposium on Field- network for EEG-Based motor imagery
Programmable Gate Arrays (2017) 5–14, classification, IEEE Trans. Neural Syst. Rehabil.
https://doi.org/10.1145/3020078.3021740. Eng. 27 (10) (2019) 2164–2177, https://doi.org/
Pr
This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4724661
signals, J. Neural Eng. 14 (1) (2016) 16003, Control Optim., vol. 11, p. 100231, Jun. 2023, doi:
ed
https://doi.org/10.1088/1741-2560/14/1/016003. 10.1016/J.RICO.2023.100231.
[25] Asama Baghadadi,Yassine Aribi Rahma [36] Mane, Megha (2023), “An EEG Recordings
Fourati,Najla Halouani, PatrickSiarry, Adel M. Dataset for Mental Stress Detection”, Mendeley
Alimi DASPS: A Database for Anxious States Data, V1, doi: 10.17632/wnshbvdxs2.1
based on psychological Stimulation, 2019
iew
http://www.regim.org/publications/databases/dasps
/
[26] M. A. Ozdemir, M. Degirmenci, O. Guren and A.
Akan, "EEG based Emotional State Estimation
using 2-D Deep Learning Technique," 2019
Medical Technologies Congress (TIPTEKNO),
Izmir, Turkey, 2019, pp. 1-4, doi:
10.1109/TIPTEKNO.2019.8895158.
v
[27] Murugappan, M. and Boon, Hui and Raj, Alex Noel
Joseph and Krishnan, Gokul and Palanisamy,
re
Karthikeyan, ECG Morphological Features Based
Sudden Cardiac Arrest (SCA) Prediction Using
Nonlinear Classifiers", Advances in Electrical and
Computer Technologies, Springer Singapore,789—
801, 2020.
[28] S. Caviola, E. Carey, I. C. Mammarella, and D.
er
Szucs, “Stress, time pressure, strategy selection and
math anxiety in mathematics: A review of the
literature,” Front. Psychol., vol. 8, no. SEP, p. 1488,
Sep. 2017, doi:
pe
10.3389/FPSYG.2017.01488/BIBTEX.
[29] H. Zhang, L. Feng, N. Li, Z. Jin, and L. Cao,
“Video-Based Stress Detection through Deep
Learning,” Sensors 2020, Vol. 20, Page 5552, vol.
20, no. 19, p. 5552, Sep. 2020, doi:
10.3390/S20195552.
[30] C. M. MacLeod, “Stroop Effect in Language,”
ot
10.1016/J.COGR.2021.07.001.
[35] S. A. M. Mane and A. Shinde, “StressNet: Hybrid
model of LSTM and CNN for stress detection from
electroencephalogram signal (EEG),” Results
This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4724661