Grenze International Journal of Engineering and Technology, June Issue

Stress Detection using CNN Algorithm

Dr. T. Guhan1, Sachin Surya.B2, Saranya.E3, Thejeswaran.S4 and Vignesh.S5
1
Associate Professor, Karpagam College of Engineering, Coimbatore, Tamil Nadu
2-5
Final year IT Students, Karpagam College of Engineering, Coimbatore, Tamil Nadu
Email guhan.t@[Link], 20f241@[Link], 20f246@[Link], 20f252@[Link], 20f257@[Link]

Abstract— Detecting stress is a critical aspect of mental health monitoring. In this project, we
explore the potential of deep learning, specifically Convolutional Neural Networks (CNNs), to
automatically identify stress levels from Electroencephalography (EEG) signal data. EEG signals,
originating from electrodes on the scalp, possess complex spatial and temporal patterns. CNNs,
known for their spatial feature extraction capabilities, offer a powerful tool for unveiling intricate
stress-related patterns. Our approach involves data pre-processing, division into training and
testing sets, and feature standardization. The custom CNN architecture, with three output units
representing distinct stress levels, achieves high accuracy in classifying emotional states. This
research highlights the applicability of deep learning in mental health, offering a data-driven
approach to timely stress detection and improved well-being. The project underscores the
growing role of deep learning in addressing mental health challenges.

Index Terms— Stress Detection, Deep Learning Algorithms, Convolutional Neural Networks,
Data Pre-processing, Electroencephalography.

I. INTRODUCTION
Stress is a complex and pervasive human experience that manifests as the physiological and emotional response
to anxiety or duress in challenging conditions. It transcends our mental and physical well-being, impacting our
core existence. Stressors, both internal and external, are catalysts for stress, ranging from professional life to
interpersonal relationships to financial concerns. The impact of stress on individuals varies, with "good" and "bad"
stress depending on how they navigate life's trials and tribulations. The human psyche is resilient and adaptable,
developing defensive mechanisms in response to prior stressful experiences. Eustress, a unique form of stress,
arises when individuals anticipate a stimulating event within their immediate surroundings, causing a noticeable
increase in pulse rate. The extent of stress extends beyond the surface, delving into the depths of our psychological
and physiological well-being. The ability to perceive and react to stressors is highly individual, with one person's
overwhelming anxiety-inducing event serving as another's stimulating challenge. The manner in which stressors
are handled, whether with resilience or vulnerability, casts a long shadow on the ultimate impact of stress.

II. PROBLEM DEFINITION

Stress is a prevalent and significant health concern affecting people worldwide, with far-reaching implications for
both mental and physical well-being. However, detecting stress levels accurately and early remains a challenging
endeavor. Stress detection is challenging due to its subjective nature and the diverse array of symptoms and

Grenze ID: [Link].10.2.399

© Grenze Scientific Society, 2024
triggers. The absence of an efficient, accurate, and accessible stress detection system hinders timely intervention
and effective stress management. Early identification and management of stress are essential to prevent the
escalation of stress-related health issues. There is a growing need for innovative solutions that leverage advanced
technologies, such as deep learning, to improve stress detection and promote better mental health outcomes.

III. METHODOLOGY
A. DATA COLLECTION
In the initial phase of our project, we collected EEG data, which is instrumental in stress detection. The data
comprises EEG signal values that provide insights into individuals' emotional states. The data collection process
involved proper equipment setup and the placement of electrodes on the scalp for signal acquisition.
B. DATA PRE-PROCESSING
To ensure the data's suitability for machine learning, we performed data pre-processing steps. This included
separating the features (EEG signal values) and the target variable (encoded emotions) denoted as 'X' and 'y',
respectively. Further, the dataset was divided into training and testing subsets using a 80-20split, with a random
seed set for reproducibility. Standardization of the features was also a critical aspect of pre-processing. The
'Standard Scaler' from the scikit-learn library was applied to normalize the EEG signal values, making them
suitable for model training.

Fig: 1.1 Dataset

Fig: 1.2 Data Cleaning Fig: 1.3 Data Pre-processing

C. DESIGN OF THE MODEL

In this module, we designed a Convolutional Neural Network (CNN) architecture tailored for the task of stress
detection from EEG data. The architecture consists of three layers: An initial dense layer with 64 units and Re LU
activation function, serving as the input 13 layer, which accepts the standardized EEG data. A second dense layer
with 32 units and Re LU activation, designed to capture hierarchical features. An output layer with 3 units and a
soft max activation function, corresponding to the three unique emotions associated with stress levels. This CNN
architecture was chosen due to its effectiveness in handling spatial data like EEG signals, allowing it to
automatically learn complex features from the input data.
D. TRAINING MODULE
The training module involves the fitting of our CNN model to the pre-processed EEG data. The training hyper
active parameters were acclimated, with 50 ages chosen as the number of training duplications, and a batch size
set at 32 for effective processing. The' adam' optimizer was employed for model training. The model was fine-

5232
tuned during this phase to learn the patterns and connections within the EEG data that are reflective of colorful
emotional countries associated with stress situations.
E. PERFORMANCE VALIDATION
In this phase, we validated the performance of our trained CNN model. The model's architecture and training were
evaluated based on key performance metrics, including accuracy, which assesses the model's ability to correctly
classify emotions related to stress. The 'validation_data' parameter was utilized to assess the model's performance
on a separate testing dataset, ensuring that it generalizes well to unseen data. The model summary was printed,
providing insights into its architecture and parameter count. This project report encompasses these four crucial
modules, illustrating the comprehensive process of stress detection using EEG data and the utilization of deep
learning techniques.

Fig:1.4 Flow Diagram of the Model

V. LITERATURE REVIEW
1. A Mental Stress detection Using Wearable sensors and machine Learning Techniques
Author: Shruti gedam
The study highlighted the importance of identifying stress stimuli and implementing early recognition techniques
in different environments such as driving, working, and studying. Machine learning classifiers like RF, SVM, and
decision trees were found to be the most effective in stress detection due to their superior performance compared
The study emphasized the significance of wearable sensors and machine learning techniques in detecting mental
stress, with a focus on developing high-accuracy models that are effective and affordable for real-time stress
monitoring. The study proposed a Multimodal Hierarchical CNN Feature Fusion approach for stress detection
using physiological signals, specifically Electrodermal Activity (EDA) and Electrocardiogram (ECG) signals. The
methodology involved combining deep and shallow features in a hierarchical feature set and utilizing multimodal
fusion at the intermediate level to enhance the efficiency of the stress detection model. By leveraging the
complementarity between high-level and low-level features through hierarchical feature fusion, the model aimed
to improve stress detection accuracy.
2. Multimodal Hierarchical CNN Feature Fusion for Stress Detection
Authors: RADHIKA KUTTALA, RAMANATHAN SUBRAMANIAN, VENKATA RAMANA MURTHY ORUGANTI
Results from the study showed that the proposed approach outperformed several classical Machine Learning (ML)
and Deep Learning (DL) techniques in stress detection. By effectively utilizing hierarchical feature fusion and
multimodal fusion, the model demonstrated enhanced performance in learning across modalities. The experimental
findings indicated the effectiveness of the methodology in utilizing the complementarity between low-level and
high-level information, leading to improved stress detection capabilities. The promising results suggested that the
proposed model could be valuable for clinical practitioners in stress detection applications.
Overall, the study highlighted the significance of hierarchical feature fusion and multimodal fusion in improving
stress detection accuracy using physiological signals, showcasing the potential for advanced models to enhance
well-being and productivity through early stress detection.
3. Stress Detection in IT Professional by Image Processing and Machine Learning
Author: [Link] Murthy
The study discussed in the PDF file focuses on stress detection in working employees, particularly in the IT
industry, using image processing and machine learning techniques. The ideology behind this method is to create a

5233
system that can automatically detect stress levels in employees based on their facial expressions captured through
images or video frames. By analyzing facial expressions, the system aims to classify stress into different levels
such as Angry, Disgusted, Fearful, or Sad. By combining image processing and machine learning, this study offers
a non-intrusive and efficient way to monitor and manage stress levels in working individuals, ultimately aiming to
improve employee well-being and productivity in the workplace.
4. Stress Detection in IT Professionals using Real Time Videos
Author: Prof. Rajitha P R
The ideology method used in the study on "Stress Detection in IT Professionals using Real Time Videos" involves
the application of Deep Learning techniques, specifically Convolutional Neural Networks (CNN), for analyzing
video frames to detect stress indicators such as facial expressions and body movements. By training the dataset
with CNN and testing the model using live streams, the researchers were able to deploy the stress detection system
using a Flask API.
The results gained from this study include the successful development of a video-based stress detection system
that can monitor IT professionals in real-time. By analyzing facial expressions, lip and eyebrow movements, and
body actions, the system can accurately identify stress levels without the need for contact sensors. This innovative
approach provides insights into stress management and employee well-being, enabling proactive measures to be
taken to address stress-related issues in the workplace.
5. Detection and Analysis of Stress in IT Professionals by Using 5ML Techniques
Author: Dr. Siddartha B K
The study aims to develop an autonomous system that can assess stress levels based on physiological signs and
machine learning algorithms . The authors mention the use of ML methods such as LR, KNN, RF, DT, and SVM
for classification, along with Python IDE and Sci-kit learn machine learning libraries. The future work section
discusses the potential of building a mobile application using an API to help employees easily check their stress
levels. The study emphasizes the harmful effects of psychological stress on health and the importance of
identifying stress in various contexts, including work environments and clinical measures.

VI. RESULT
The proposed system for stress discovery using EEG signals using CNNs is a machine literacy- grounded approach
that utilizes the power of convolutional neural networks to directly and efficiently classify EEG signals into stress
and non- stress orders. The frame suggests a new way of EEG( Normal/ Abnormal) bracket first, capture the idle
confederations of actors by rooting stress vaticination grounded on network connectivity, and next, apply extant
data mining ways to bracket grounded on the uprooted vaticination. In the original study, modularity maximization
was employed to prize stress vaticination. The superiority of this frame over other representative relational literacy
styles has been vindicated with stress vaticination stress data. we proposes an effective edge centric approach to
prize meager stress vaticination. Stress is a common problem affecting numerous people in moment's fast- paced
society. Electroencephalography( EEG) is an on-invasive fashion used to record electrical exertion in the brain
and has been set up to be useful in studying stress. In this study, we aim to classify stress situations using CNN
algorithms grounded on 5 EEG signals.

Fig: 1.5

5234
VII. CONCLUSION
In this project, we have successfully demonstrated the potential of deep learning, specifically Convolutional Neural
Networks (CNNs), in the domain of stress detection using EEG signal data. Stress, a critical factor in mental health,
demands precise and timely identification, and our approach addresses this challenge effectively. The CNN-based
model, trained on EEG data, showcases remarkable accuracy in classifying various emotional states related to
stress, emphasizing the feasibility of automated stress detection. By combining the power of CNNs, data pre-
processing, and appropriate training, we have created a valuable tool for early intervention and improved mental
well-being. The implications of this research extend beyond the immediate scope of stress detection. The project
exemplifies the broader application of deep learning in mental health, emphasizing the significance of data-driven
approaches. It underscores the potential to transform and enhance mental well-being through advanced technology.
As the field of mental health continues to evolve, our work contributes to the on going mission of fostering early
intervention and improved mental health outcomes.

REFERENCES
[1] Shruti gedam,s Sanchitha paul,” A Mental Stress detection Using Wearable sensors and machine Learning Techniques”,
Institute of Electrical and Electronics Engineers Sensors (IEEE), VOL. 9, JUNE 2, 2021
[2] RADHIKA KUTTALA , RAMANATHAN SUBRAMANIAN , VENKATA RAMANA MURTHY ORUGANTI,”
Multimodal Hierarchical CNN Feature Fusion for Stress Detection ”, Institute of Electrical and Electronics Engineers
Sensors (IEEE), VOLUME 11,2023.
[3] Chinchinada Kesava Sree Manojna, [Link] Murthy,” STRESS DETECTION IN IT PROFESSIONAL BY
IMAGE PROCESSING AND MACHINE LEARNING” Engineering Sciences, Vol 13 Issue 07,2022
[4] Aleena Ann, Prof. Rajitha P R, Dr.T. Mahalekshmi ,” Stress Detection in IT Professionals using Real Time Videos”,
International Journal of Research Publication and Reviews, Vol 2, no 12, pp 788-794, December 2021
[5] Bindu K N, Dr. Siddartha B K, Dr. Ravikumar G K,” Detection and Analysis of Stress in IT Professionals by Using 5ML
Techniques”, International Journal for Research in Applied Science & Engineering Technology (IJRASET) , Volume 10
Issue VII July 2022

5235