Stress Monitor Using CNN

Mr.M.KISHORE BABU
Assistant Professor, Department of Computer Science and Engineering.

Pranathi Gunnam1, Munni Karavalla2,Tulasi Gurram3, Yaswanth sai Madala4, Saikrishna Kancharlapalli5
1
Dept. of Computer and Science engineering, VVIT, Guntur, India
2
Dept. of Computer and Science engineering, VVIT, Guntur, India
3
Dept. of Computer and Science engineering, VVIT, Guntur, India
4
Dept. of Computer and Science engineering, VVIT, Guntur, India

Abstract—Stress at work has become a serious issue that affects such as furrowed brows, tense jawlines, and strained eyes, serve
worker health and business success. Traditional methods of as reliable indicators of stress [4][6]. By leveraging these
measuring stress, such as self-reports and surveys, are not always insights, deep learning models like MobileNetV2 and
reliable and may not provide immediate feedback. To overcome
these problems, this paper proposes a real-time stress monitoring Convolutional Neural Networks (CNNs) have been employed
system that analyzes facial expressions and detects stress using to process facial data efficiently, ensuring high accuracy and
CNNs. The system is suitable for use in modern workplaces adaptability across diverse workplace scenarios [1][7]. Such
because it uses MobileNetV2 for fast and scalable processing. It systems have proven to be more effective than traditional
also features a chatbot that is powered by an artificial neural methods, providing both real-time and continuous monitoring
network (ANN) that provides customized stress- reduction
recommendations, including relaxation techniques and counseling of stress levels without requiring any physical contact with the
materials. Based on the results of the pilot test, the system is user.
accurate and efficient, thus making it a useful tool for managing Moreover, physiological measures such as body tempera-
stress in different work settings. ture, skin conductance, and heart rate can also be helpful
Index Terms—Stress detection, Convolutional Neural Networks in stress detection. However, these measures often require
(CNN), MobileNetV2, Artificial Neural Networks (ANN), Real-
time monitoring. wearable technology, which may be inconvenient or unfeasible
in an office environment [2][3]. By focusing on facial emotion
recognition, our approach addresses this challenge by offering
I. INTRODUCTION
a non-intrusive and user-friendly way of stress monitoring
Stress at work has become a major issue in today’s fast- without requiring additional sensors or hardware.
paced, fiercely competitive world, especially in the IT industry Our project advances over the previous ideas of techno-
where performance and creativity are valued highly. The logical progression through the building of a real-time stress-
mental and physical health of employees is greatly impacted by detection system combined with personalized intervention and
this stress, which is caused by elements like heavy work- loads, facial emotion recognition. Here, the system applies contin-
strict deadlines, and job uncertainty. The World Health uous monitoring with the help of the web camera mounted
Organization (WHO) claims that stress is a major contributor on the wall or placed inside a cubicle. The expressions are
to mental health issues and can result in serious health analyzed against CNNs and compared to stress patterns, mak-
issues like depression, diabetes, and cardiovascular illnesses. ing the system precise while scalable for a modern workplace
Conventional methods of stress detection are still insufficient, setting. In addition to detecting stress, the system integrates
despite organizational initiatives to promote mental health a chatbot powered by an artificial neural network (ANN) to
advantages. Conventional techniques for measuring workplace provide employees with tailored stress management tips and
stress, such self-reported questionnaires and surveys, are labo- resources. This personalized support enhances the effective-
rious, subjective, and don’t offer useful, immediate insights. ness of the system, enabling employees to proactively manage
According to the literature, these approaches frequently lack their stress and improve their mental well-being.
precision and might not reflect the dynamic nature of stress By combining state-of-the-art AI techniques, such as
experienced by employees during their workday [5][6][7]. MobileNetV2 for efficient facial expression recognition and
Moreover, manual assessments require substantial effort, both ANN for personalized recommendations, this system has the
from administrators and participants, leading to inefficiencies potential to transform workplace stress management. It
and a lack of timely intervention. promises to provide real-time feedback and actionable insights,
The combination of machine learning (ML) and artificial helping organizations foster healthier work environments and
intelligence (AI) technologies has become a ground-breaking improve employee productivity and morale.
way to overcome these constraints. These advanced tools offer
innovative, non-intrusive methods for detecting stress in real
time. Specifically, computer vision techniques have proven II. LITERATURE SURVEY
effective in analyzing facial expressions to identify stress levels 1) Stress Detection Using Machine Learning and Image
accurately. Research has shown that micro-expressions, Processing
 a web interface for user registration, live detection, and secure
data handling. By automating stress detection through
photographs and reducing dependence on questionnaires, this
system offers a scalable solution for timely stress intervention
in workplace environments.

4) Detecting stress and anxiety via facial clues

Authors:G. Giannakakis, et al
Using video-recorded face clues, this literature review
establishes a model for detecting and identifying stress/anxiety
emotional states. Through some kind of range of external
and internal stresses, a complete experimental method was
designed to induce systematic diversity in emotional states
(neutral, relaxed, and stressed/anxious). In order to evaluate
emotion expression more clearly and accurately, the paper
concentrated mostly on non-voluntary and semi- voluntary
Fig. 1. Block Daigram
facial cues. Sight related activities, oral activity, gesture
recognition characteristics, and heart rate were also examined
Authors:E.Padma, Talapaneni Praveen, Shaik Karimulla-
This study aims to address stress detection and management 5) Stress Prediction in the Workplace Using K-Nearest
using machine learning and image processing techniques. Neighbors (KNN)
The system processes facial expressions captured via Authors:Dr.R Guru,Prajwal Annigeri,Nagaraj Huddar, Kiran
cameras, analyzes stress indicators like anger or sadness KumarH,Prajwal KP -
using K-Nearest Neighbor (KNN) classifiers, and generates This research focuses on workplace stress detection,
reports on stress levels. It incorporates modules such as particularly in IT sectors, using the KNN algorithm. Key
user management, data preprocessing, and machine learning. factors like gender, age, family issues, and workplace dynamics
The existing systems, which rely on intrusive sensors, were analyzed to classify stress levels into four categories. The
are improved upon by making the approach non-intrusive system achieved an 87.22% classification accuracy and
and efficient, offering continuous monitoring and stress provided recommendations to mitigate stress. Real-time
management. Future enhancements include integration with usability and a user-friendly GUI were incorporated, making it
biomedical IoT sensors and extending features to analyze a practical tool for improving workplace mental health and
physiological and behavioral data, making it a scalable productivity.
solution for workplace mental health improvement.
2) Stress Detection Using Machine Learning and Deep 6) Stress Detection Using Machine Learning
Learning Authors:Prof.Rohini Hanchate, Harshal Narute, Siddharam
Authors:Z.Zainudin, S.Hasan, S.M.Shamsuddin and S.Argawal Shavage, Karan Tiwari-
This work employs both machine learning and deep learning to This study introduces a system that utilizes machine learning
classify stress levels using bio-signals like Galvanic Skin and image processing techniques to detect and manage stress
Response (GSR) and Electrocardiogram (ECG). Data from 252 levels in individuals. The proposed model incorporates a
participants was processed, extracting key features such as live camera and employs algorithms such as Convolutional
mean and standard deviation. Algorithms like Decision Tree Neural Networks (CNN) and Haar Cascade to analyze facial
(DT), KNN, and deep learning models were evaluated, with expressions and accurately identify stress patterns. By
DT achieving the best performance (75% accuracy). The preprocessing captured images, extracting key features, and
findings highlight the importance of data preprocessing and classifying stress levels, the system provides users with stress
algorithm selection. Future work suggests incorporating more percentage results and recorded facial expressions to help
classifiers and extending datasets to improve accuracy, identify potential triggers. The research highlights the
emphasizing early stress detection’s relevance in health importance of automating stress detection as a reliable and
monitoring. objective alternative to traditional self-reported methods and
demonstrates the effectiveness of combining physiological
3) Stress Detection in IT Professionals Using Image signal analysis with computer vision to tackle the growing issue
Processing and Machine Learning of stress in various environments.
The project aims to detect mental and physical stress among IT
professionals by analyzing facial expressions. It integrates 7) Image Processing and Machine Learning Techniques
advanced machine learning and image processing techniques, used for Stress Detection
improving upon traditional methods by providing real-time Authors: Nisha Raichur, et al.
analysis alongside survey data for insights. Features include A real-time non-intrusive video is produced in this paper,
which detects by analyzing a person’s facial expression, can IV. PROPOSED SYSTEM
determine their emotional state. Each video frame contains The proposed system utilizes an advanced architecture of
an individual feeling, and the stress level is determined in the Convolutional Neural Network (CNN) to allow for real-time
hours after the video capture[18]. A mechanism is applied that stress detection based on facial emotions. Unlike traditional
enables us both to train a system and analyze feature prediction approaches, which rely on manual surveys or algorithms like
differences. The paper findings suggest that the developed SVM and logistic regression, this system provides an auto-
technique works effectively with a generic model of all ages mated, non-intrusive method for identifying stress levels. Us-
ing a webcam, the system captures employees’ facial images,
8) Analysis of Stress Detection among Employees which are then processed through CNN layers to extract key
usingMachine Learning Techniques features indicative of stress. These features are classified into
Authors:B. K. Kiranashree, et al categories such as Angry, Disgusted, Fear,Happy, Sad,Surprise
This research focuses on using Artificial Neural Network and Neutral. The detected emotion is above a bounding box that
(ANN) classifiers to detect stress levels in employees. highlights the face. This does not interfere with the employees
Employees’ mental state is assessed by questions that use work flow. It is seamless and non interruptive of the correct
the Perceived Stress Scale, which also includes their health and detection of stress.
wellbeing by measuring cardiac fluctuation and blood pressure. CNN model consists of different layers which include the
Because an ANN classifier can learn a large amount of data, we convolutional, pooling, fully connected layers to extract spatial
believe it will produce a rather more accurate solution . features, reduce dimensionality, and identify stress very
accurately. It uses the Adam optimizer and categorical cross-
entropy loss function, so it is strong and efficient. The system
III. EXISTING SYSTEM adopted the MobileNetV2 architecture. It is light in weight and
The existing system for stress detection focuses on recog- has efficient design that includes depthwise separable convolu-
nizing facial emotions but does not provide specific stress-level tions to minimize the overhead of computations. Therefore, it
analysis. It utilizes the K-Nearest Neighbor (KNN) algorithm, can be deployed on low resource devices. Transfer learning
which is slow in processing, consumes more memory, and is used to further improve the performance of the system,
is computationally expensive, especially for handling large enabling it to adapt well to workplace-specific datasets and
datasets. This approach is limited in its capabilities, as it achieve robust generalization.
lacks real-time stress detection and does not include features In addition to stress detection, the system includes a per-
like personalized advice or suggestions for stress management. sonalized chatbot powered by an Artificial Neural Network
Moreover, the system’s reliance on manual methods, such (ANN), providing employees with tailored stress management
as questionnaires, makes it less efficient for continuous and tips, mental health resources, and interactive counseling sug-
automated stress monitoring. gestions. The system’s performance is evaluated using metrics
Additionally, the existing system is not scalable and strug- such as accuracy, precision, recall, specificity, F1-score, and
gles with handling large volumes of data effectively. It does not false positive rate, ensuring its reliability and effectiveness. Its
integrate advanced techniques for accurate and robust stress versatile design makes it suitable for applications in corporate
detection, resulting in limited practical application in environments, educational institutions, and healthcare settings.
workplace settings. This creates a gap in providing timely By combining real-time stress monitoring with actionable
interventions to address employee mental health challenges. support, the proposed system offers a scalable and efficient
Furthermore, the absence of advanced algorithms and features solution to workplace stress management, fostering healthier
like stress prediction models hinders its reliability in diverse and more productive environments.
workplace scenarios. As a result, the system fails to meet
modern workplace demands for proactive and efficient stress A. Advantages of this approach
management solutions. • The system uses CNN for real-time facial expression
analysis, enabling instant stress detection and bypassing
A. Disadvantages of the Existing System
delays of traditional surveys.
• The existing system detects facial emotions but lacks • MobileNetV2 and optimized image processing minimize
stress-level analysis, making it ineffective for targeted image capture needs, saving time and resources.
stress management.
• It uses the K-Nearest Neighbor (KNN) algorithm, which
B. Dataset
is slow to process data and inefficient for real-time stress The FER-2013 dataset is a widely used collection of images
detection. designed to help train and evaluate models for facial emotion
• The KNN algorithm is computationally expensive, recognition tasks. It contains more than 35,000 grayscale
memory-intensive, and faces storage issues when dealing images, each with a resolution of 48x48 pixels. These images
with large datasets. have been pre-processed to ensure that the faces are centered
and uniformly scaled, allowing for consistent analysis during
training. The dataset categorizes facial expressions into seven
 Fig. 5. Block Daigram For Chatbot

Fig. 2. Indication of stressed emotions states.

become a benchmark for developing and testing algorithms
in the field of facial emotion analysis.

V. METHODOLOGY
A. Convolutional Neural Network (CNN):
The CNN algorithm, or Convolutional Neural Network al-
gorithm, is a type of deep learning algorithm used in computer
vision and image processing. It uses filters or kernels to extract
features from images by performing convolution operations.
These features are then passed through a series of layers,
including pooling and activation layers, to create a feature map.
The feature map is then flattened into a vector and passed
through fully connected layers to make predictions about the
image. CNNs are used in many applications, such as ob-
ject detection, facial recognition, and medical imaging. CNN
(Convolutional Neural Network) can play a critical role in
object detection and image classification tasks in the proposed
system, particularly in computer vision applications. CNNs are
known to be highly effective in recognizing visual patterns
Fig. 3. Indication of not stressed emotions states in images and detecting objects with high accuracy. They
can extract relevant features from images and classify them into
different categories, making them useful in applications such as
emotional classes: Angry, Disgust, Fear, Happy, Sad, Surprise, facial recognition, object detection, and autonomous vehicles.
and Neutral. With 28,709 images allocated for training and
3,589 for testing, it provides a comprehensive foundation for B. MobileNetV2 Architecture
building and testing emotion recognition systems. To improve MobileNetV2 employs depthwise separable convolutions,
the model’s ability to generalize to real-world scenarios, reducing computational complexity while maintaining high
data augmentation techniques such as flipping, rotation, and accuracy. The architecture is optimized for deployment on
cropping are commonly applied. These methods increase the resource-constrained devices
diversity of the dataset, enabling the model to perform more
reliably under varying conditions. The FER-2013 dataset has C. Chatbot Design
An ANN is used for intent recognition and response gen-
eration. The chatbot provides real-time, personalized recom-
mendations for stress management based on detected stress
levels.
VI. EXPERIMENT AND ANALYSIS
After logging in, users can access the Real-Time Emotion
Detection feature. The system prompts the user to enter their
Fig. 4. Emotions name and start the real-time recording. Using a live camera,
 Fig. 6. Main page

Fig. 8. Uploaded Video or Image

Fig. 7. Take a Test

the system captures the user’s facial expressions and processes

them in real time. It identifies emotions such as Angry, Sad,
Happy, or Neutral and overlays the detected emotion on a
bounding box displayed over the user’s face. The detected
emotion is immediately presented to the user as a prediction Fig. 9. Using Live Cam Result
result, providing real-time insights into their stress level while
maintaining high accuracy and responsiveness.
Additionally, the system provides an option for File Upload system processes the input to detect relevant features. If the
Emotion Prediction. Users are required to enter their name and input is an image, the user uploads it, and the system analyzes
upload either an image or a video file for emotion analysis. The it for faces. In the case of live camera use, the system
system accepts the file through a ”Choose File” button, continuously processes the video feed to detect emotions in
ensuring that only valid formats (e.g., JPG for images and MP4 real-time. This makes it possible to detect and analyze facial
for videos) are allowed. Upon successful upload, the user clicks expressions or emotional states from either a still image or
the ”Predict” button to process the file. The system analyzes the moving visuals.
content to detect and display the associated emotions, Once the system completes its analysis and detects the
presenting the prediction result along with details about stress emotion, the result is displayed above the bounding box. This
levels. This feature offers users a convenient and detailed ensures the output is easy to see in context with the area being
evaluation of their emotional state through uploaded media. If analyzed. The result is shown in a clear format, including
the user clicks on the Image Page, they can upload an image for a timestamp, the file name, and the predicted emotion. The
analysis and view the results. Additionally, users can access timestamp marks the exact moment the prediction was made,
previously uploaded images and their corresponding results. while the file name helps identify the specific image or video
This functionality is particularly valuable for future medical processed. Finally, the predicted emotion, such as stress or not
references, enabling psychiatrists to analyze historical data and stressed, is determined based on various parameters that will be
provide personalized assistance to the user. trained in the system.
If the user clicks on the Live Cam option, the system acti-
vates the camera and begins monitoring the face of the person
in front of it. The system processes the facial expressions in real
time and displays the detected emotion and stress level.
Furthermore, the system is capable of monitoring multiple
faces simultaneously, making it suitable for group scenarios or
environments requiring comprehensive stress monitoring.
After uploading an image or using the web camera, the Fig. 10. Result
 Metric Value
Accuracy 0.9084
Precision 0.9188
Recall 0.8978
Fig. 11. Confuse Matrix F1-Score 0.9082

Class Precision Recall F1-Score Support

Anger 0.82 0.84 0.83 828
Disgust 0.99 0.96 0.97 841
Fear 0.90 0.91 0.90 836
Happy 0.84 0.88 0.86 833
Neutral 0.70 0.75 0.72 806
Sad 0.72 0.68 0.70 798
Surprised 0.85 0.80 0.83 830
Accuracy 0.83 0.83 0.83 (5772)
Macro average 0.83 0.83 0.83 5772
Fig. 12. Graphs Weighted average 0.83 0.83 0.83 5772

VII. RESULT
VIII. CONCLUSION
A. performance Evaluation The CNN-based stress monitoring system offers a non-
intrusive, real-time solution, overcoming the limitations of
Performance evaluation is crucial for understanding how traditional methods. Its integration with MobileNetV2 ensures
effective machine learning models are. By using various met- efficiency and scalability for modern workplaces. The ANN-
rics, cross-validation techniques, and confusion matrices, prac- based chatbot provides personalized stress management tips,
titioners can gain valuable insights into model performance. empowering proactive stress management. Future research
This knowledge helps them make informed decisions about could incorporate physiological data and expand to mobile and
model selection and identify areas that need improvement. wearable devices, enhancing overall employee well-being.
Ultimately, this process leads to the development of robust and
reliable machine learning applications.
A confusion matrix is a performance measurement tool used IX. REFERENCES
for classification problems in machine learning. It is a table that [1]E. Padma, T. Praveen, and S. Karimulla, ”Stress Detec-
describes the performance of a classification model by tion Using Machine Learning and Image Processing,” Journal
comparing the actual target values with the values predicted by of Engineering Sciences, vol. 13, no. 3, pp. 417–422, Mar.
the model. 2022, ISSN: 0377-9254.
True Positives (TP): The number of instances correctly [2]Zainudin Z Shamsuddin S M and Hasan S 2019 The
predicted as positive. True Negatives (TN): The number of International Conference on Advanced Machine Learning
instances correctly predicted as negative. False Positives (FP): Technologies and Applications (AMLTA2019) 921 43-51
The number of instances incorrectly predicted as positive (Type [3][Authors Not Provided], ”Stress Detection in IT Pro-
I error). False Negatives (FN): The number of instances fessionals Using Image Processing and Machine Learning,”
incorrectly predicted as negative (Type II error). [Journal Name], vol. XX, no. XX, pp. XX–XX, Year.
 [4] G. Giannakakis, D. Manousos, F. Chiarugi, “Stress ence on Computer Communication and Informatics (ICCCI),
and anxiety detection using facial expressions from videos,” 2020, pp. 1-7, doi: 10.1109/ICCCI48352.2020.9104113.
Biomedical Signal processing and Control”, vol. 31, pp. 89 - [18]U. S. Reddy, A. V. Thota and A. Dharun, ”Machine
101, January 2017. Learning Techniques for Stress Prediction in Working Employ-
[5]R. Guru, P. Annigeri, N. Huddar, K. Kumar, and P. K. P., ees,” 2018 IEEE International Conference on Computational
”Stress Prediction in the Workplace Using K-Nearest Neigh- Intelligence and Computing Research (ICCIC), 2018, pp. 1-4,
bors (KNN),” International Research Journal of Modernization doi: 10.1109/ICCIC.2018.8782395.
in Engineering, Technology and Science, vol. 4, no. 7, pp. [19] Sharma, L.D. , Bohat, V.K. , Habib, M. “Stress Detec-
1982–1989, Jul. 2022. tion Using EEG Signals”, Journal of Advanced Research in
[6]R. Hanchate, H. Narute, S. Shavage, and K. Tiwari, Dynamical and Control Systems,2017.
”Stress Detection Using Machine Learning,” International Re- [20] Liu, D., Ulrich, M.. Listen to your heart: Stress predic-
search Journal of Engineering and Technology (IRJET), vol. tion using consumer heart rate sensors 2015.
10, no. 5, pp. 1–7, May 2023.
[7] Sriramprakash.S , Prasanna Vadana. D,O.V.Ramana
Murthy ”Stress Detection in Working People” 7th International
Conference On Advances In Computing and Communica-
tions,ICACC 2017, August 22-24, 2017.
[8] Nisha Raichur, Nidhi Lonakadi, Priyanka Mural, “Detec-
tion of Stress Using Image Processing and Machine Learning T
echniques”, vol.9, no. 3S, July 2017.
[9] Ankita Patil, Rucha Mangalekar, Nikita Kupawdekar,
Viraj Chavan, Sanket Patil, and Ajinkya Yadav, “Stress Detec-
tion in IT Professionals By Image Processing And Machine
Learning,” International Jo urnal of Research in Engineering,
Science and Management, Vol.3, No.1, January-2020.
[10] Mrs. Megha V Gupta Dr. Shubhangi Vaikole “ Recog-
nition of Human Mental Stress Using Machine Learning
Paradigms”,2020.
[11] U. S. Reddy, A. V. T hota and A. Dharun, ”Machine
Learning T echniques for Stress Prediction in Working Em-
ployees,” 2018 IEEE International Conference on Computa-
tional Intelligence and Computing Research (ICCIC), Madu-
rai, India, 2018.
[12] Huijie Lin, Jia Jia, JiezhonQiu, “ Detecting stress based
on social interactions in social networks”, IEEE T ransactions
on Knowledge and Data Engineering, 2017.
[13] Koldijk, S., Neerincx, M.A., Kraaij, W.. Detecting work
stress in offices by combining unobtrusive sensors. IEEE T
ransactions on Affective Computing 2016.
[14] Pramod Bobade, Vani M : Stress Detection with Ma-
chine Learning and Deep Learning using Multimodal Physi-
ological Data Proceedings of the Second International Con-
ference on Inventive Research in Computing Applications
(ICIRCA-2020) IEEE Xplore Part Number: CFP20N67-ART;
ISBN: 978-1-7281-5374-2.
[15] Stress Detection Using Machine Learning and Image
Processing Mahesh Kanthale1, Akshata Shendkar2, Payal
Kadadhekar3,Siddheshwari Rao4,Prof. Dheeraj Patil5.
[16] Stress detection using natural language processing and
machine learning over social interactions Tanya Nijhawan1 ,
Girija Attigeri2 and T. Ananthakrishna1: Nijhawan et al. Jour-
nal of Big Data (2022) 9:33 https://doi.org/10.1186/s40537-
022-00575-6.
[17]S. Monisha, R. Meera, V. Swaminath R. and A. Raj
L., Predictive Analysis of Student Stress Level Using Na¨ıve
Bayesian Classification Algorithm, 2020 International Confer-