Proceedings of the 7th International Conference on Computing Methodologies and Communication (ICCMC-2023)

IEEE Xplore Part Number: CFP23K25-ART; ISBN: 978-1-6654-6408-6

Detection of Stress by Machine Learning in IT

Industry
Hari haran S, Budugarla Rohith B,
M ohana Priya D, Computer Science and Engineering, Computer Science and Engineering,
Computer Science and Engineering, Sri Eshwar College of Engineering, Sri Eshwar College of Engineering,
Sri Eshwar College of Engineering, Coimbatore, India, Coimbatore, India,
Coimbatore, India, [email protected]. [email protected].
[email protected].
Jagadeesh kumar P,
2023 7th International Conference on Computing Methodologies and Communication (ICCMC) | 978-1-6654-6408-6/23/$31.00 ©2023 IEEE | DOI: 10.1109/ICCMC56507.2023.10083884

Hariprasath M , Computer Science and Engineering,

Computer Science and Engineering, Sri Eshwar College of Engineering,
Sri Eshwar College of Engineering, Coimbatore, India,
Coimbatore, India, [email protected].
[email protected].

Abstract—The primary goal of our design is to reduce a threshold figure for determining the level o f stress is
stress among IT professionals by combining graphical calculated from the data from every sensor.
technology literacy and image processing techniques. Our
system is a continued improvement over the old stress To estimate the stress levels, the heart rate
discovery systems that tried to prevent live discovery and statistics, GSR, and frequency domain facets of the heart
specific comforting, but this system includes periodic worker rate and its variability (HRV) in the ECG were examined. In
analysis and live invention, which helps it recognize workers order to further quantify an xiety levels, many causes are
who really are enduring both physical and compressive load asserted from how often physiological signals, such as ECG,
difficulties and to propose them appropriate stress-
EM G, GSR, BVP, etc., are used and monitored using the
management strategies by outfitting periodic check forms. Our
approach places a significant emphasis on stress management, pulse sensor. In, it is found that, by utilizing the General
creating a work environment that is stress-free for the Regression Neural Network (GRNN) model, s maller
employees, and relieving their tension while they are at work. clusters lead to better balance in stress detection. This has
the effect of making various combinations of the sensor
Keywords—Stress prediction, Boosting, Bagging, signal components that have been retrieved more effective at
Decision Trees, Healthcare, Machine Learning predicting the continuous anxiety level. Examp les of
frequency domain values entail HF power, LF power, and
I. INT RODUCT ION LF/HF. For subsequent real-time stress detection, the
Detection of stress levels involves the use of corporeal do main attributes of the heart signal, such as the
positive coping systems that disturb our socioeconomic way mean, median, and standard deviation, are factored. Two
of life. According to the World Health Organization stressors are involved in the grading approach or use of a
(WHO), one in four people suffer fro m the mental health decision tree as PLDA: the picking task and the stoop-based
issue of stress. Hu man stress causes mental and word test. The classification based on pressures is
socioeconomic issues, such as a loss of focus at work, inadequate, per the authors. Using data from the HRV and
strained relationships with co workers, despair, and, in the ECG signals, Gorski et al. developed classifiers for stress
worst circu mstances, suicide. This necessitates the provision detection in 2016. The GRNN model is used to analyses the
of counselling to help those under stress manage their stress. ECG's features and determine how much stress is present.
Hu man stress causes mental and socioeconomic issues, such Heart rate variab ility (HRV) and RR interval specifications
as a loss of focus at work, strained relationships with are used to analyses the level of stress. The scalability and
coworkers, despair, and, in the wo rst circu mstances, suicide. solid technical foundation of the Support Vector Machine
This necessitates the provision of counselling to help those feature set led to its selection. After applying various kernels
under stress manage their stress. This opens the door to the to build a series using SVM, a linear SVM wo rked better
need for a scientific instrument that uses physiological data than other more co mplicated systems on both HRV features
to detect stress levels in people. Nu merous literary works and ECG frequency characteristics. These days, new
tackle stress detection since it significantly improves technologies and goods are being introduced to the market
people's quality of life and contributes to society. Ghaderi et by IT businesses, giving the industry a fresh look.
al. concluded that features related to the respiratory process
In this investigation, it was also found that
are significant in stress identification after analy zing and
emp loyee stress levels were very high. Although many
utilizing stress. A study to predict stress levels only fro m
companies offer their staff programmers related to mental
electrocardiograms was proposed by David Liu and
health, the issue is still far fro m under control. This study
colleagues (ECG) to evaluate the effectiveness of
also attempts to identify stress patterns of working
mu ltimodal sensors in identifying wo rkp lace stress.
emp loyees. Co mpanies can evaluate stress patterns and
Continuous stress levels are assessed using physical sensors
identify the elements that have a significant impact on
fro m the GSR, EM G, HR, and breathing. Stress can also be
mach ine learn ing and image processing algorithms that
measured with sensors for the heart rate, skin conductance
influence stress levels. Stress is categorized using machine
level (SCL), and mouth EM G. Multip le pattern recognition
learning techniques like KNN classifiers. The camera
methods enable automated stress detection. A tension index,
captures the image of the employee and inputs it into the

978-1-6654-6408-6/23/$31.00 ©2023 IEEE 53

Authorized licensed use limited to: BMS Institute of Technology. Downloaded on November 23,2023 at 02:41:02 UTC from IEEE Xplore. Restrictions apply.
 Proceedings of the 7th International Conference on Computing Methodologies and Communication (ICCMC-2023)
IEEE Xplore Part Number: CFP23K25-ART; ISBN: 978-1-6654-6408-6

initial stage of image processing for detection. Image least limit their use if they are concerned about becoming
processing involves transforming an image into digital form stressed out at a specific point. It is vital fo r a person's safety
and conducting major procedures on it in order to create an to take care of their health when using a computer fo r an
improved image or ext ract some key data fro m it. Video extended time. This technique uses real-time, non-intrusive
frames are used as the input, and the output is either a visual video to monitor and record a person's emotional state
or the equivalent of two of those images. Basically, an through facial expression analysis. Every v ideo frame
image includes: the image, not with tools for image capture carries a d istinct feeling that can be recognized, and finally
but instead with analysis. Machine learning, a sort of settle on the level of an xiety a few hours after the video was
artificial intelligence, offers devices the ability to shot. A novel method has been applied to develop a model
automatically learn fro m and develop based on their and track variations in feature estimates. A Python
personal experiences without being explicitly p rogrammed framework named Theano tries to decrease the execution
(AI). Machine learning builds software programs that can time as well as the development time of the deep learning
access data and use it to educate themselves. Using machine algorith m used here, the linear regression model. The
learning, exp licit programming uses "training data" to create experimental find ings indicate that the new method operates
a mathematical model. To execute the task based on satisfactorily with data using the typology of all ages.
predictions or conclusions. To discover hidden patterns that
are only barely visible in movies, image min ing is used to
C. Machine Learning Methods for Predicting Stress in
extract hidden data, associate image data, and help detect
the Workplace
patterns. Images in this topic make the co mplex
relationships between disciplines of image processing, data Today's working IT professionals frequently deal
mining, machine learn ing, and datasets visible. Data min ing, with tasks assigned to them. The risk of stress among
mach ine learning, image processing, and datasets are all emp loyees had improved because they had an improved
related disciplines that have been studied in this area. lifestyle and needed to work. Most businesses and industries
Medical publicat ions provide cautious estimates that stress offer mental health plans and attempt to identify stress
contributes to between 50 and 80 percent of all medical patterns among emp loyees that have the greatest effect on
ailments. Stress is thought to be the leading cause of their stress levels. In this respect, data from working
cardiovascular disease. Stress can increase the risk of professionals in the tech industry's responses to the OSMI
developing diabetes, ulcers, asthma, migraines, skin mental health survey from 2017 was considered. After
conditions, epilepsy, and erectile dysfunction.
comprehensive cleaning, a range of datasets were used to
II. LITERATURE SURVEY train our model. The accuracy of the previous methods was
validated by reference. Among the models tested, boosting
A. Detecting stress and anxiety using video facial
had the lowest error. A decision tree found that family
clues
history, gender, and access to health benefits at work are
In this study, face signals captured on video are
helpful factors that influence stress. With the help o f this
used to identify and analyze stress and anxiety emotions.
research, firms may now focus on ways to make their
Through a variety of internal and external stresses, a
complete experimental technique was constructed to create workplaces less stressful and happier.
systematic diversity in emotional states. The research
primarily concentrated on involuntary and semi-voluntary D. Acute stress is classified using sternal ECG-
facial cues in order to assess emotion representation more derived linear and nonlinear heart rate variability analyses
accurately. The features under research were mouth activity,
head motion parameters, eye-related events, and heart rate Recognizing chronic stress is key to predicting
estimated by camera-based photoplethysmography. In each and limit ing the probability of developing card iovascular
experimental phase, a method of feature selection has been disease. This study is a pilot study that seeks to develop a
used for selecting the most consistent features, then way to detect brief psychophysiological changes using the
classification algorith ms were applied to determine between characteristics of heart rate variability. Th is pilot study's
stress/anxiety and neutral states with reference to a relaxed objective is to identify a group of indicators and build
condition. Moreover, a ran king t ransformat ion was perceptions that may be used to identify alterations in the
suggested using self-reports to investigate the link between psychophysiology of prolonged stress. This research work
participants estimated levels of stress or anxiety and face has identified four different types of highs that each patient
features. Using the findings, certain facial cues that are experiences and codified them based on the linear and
extracted fro m mouth, eye, and head movements as well as nonlinear HRV p roperties of electrocardiograms obtained
heart activity calculated using a camera may help recognize using a patch wireless device.
the difference between stress and anxiety with good
accuracy . E. Mood recognition at work utilizing cellphones and
wearable sensors for a healthy work place:
B. Stress Detection by Image Processing and Machine Stress, anxiety, and depression in the workplace
Learning methodologies are detrimental to hu man health and productivity and have
People experience stress when they work on a significant financial implications. Recent research in this
computer for several days at a time. Even when using a area has focused on the use of sensor technologies,
computer for work cannot be totally avoided, one can at including smartphones and wearables embedded with

978-1-6654-6408-6/23/$31.00 ©2023 IEEE 54

Authorized licensed use limited to: BMS Institute of Technology. Downloaded on November 23,2023 at 02:41:02 UTC from IEEE Xplore. Restrictions apply.
 Proceedings of the 7th International Conference on Computing Methodologies and Communication (ICCMC-2023)
IEEE Xplore Part Number: CFP23K25-ART; ISBN: 978-1-6654-6408-6

physiological and movement sensors. This research work  The physiological signals stress index is
explores the possibility of using such devices for mood given by the collected features. The
recognition, focusing on work environments. Framework commonly used peak j48 method is used
that can identify five intensity levels for eight different to directly evaluate the ECG signal.
types of mood every two hours. Further, this study presents
a smartphone app (Healthy Office), designed to facilitate  It might be helpful to identify a pattern to
self-reporting in a structured manner and provide our describe the stress emotion, for various
model with ground truth. We evaluate our system in a people may react or express themselves
small-scale user study where wearable sens ing data is differently when under stress. Algorithm:
collected in an office environ ment. The proposed Bayesian Network, J48
experiments show promising results, that allows to reliably
recognize various classes of perceived moods. IV.METHODOLOGIES

F. Automated facial action unit model evaluation Image Preprocessing: H (a, b) = α. Gain and bias
during stress detection parameters, generally called M (a, b) +, > 0, are used to
depict and contrast the image. Here, an output image pixel is
Examining several facial borders can help identify H (a, b), and the input image pixel is M (a , b).
a pressure point near your home. It focuses on the Transformat ion of dots: To assess pixel values, image
automatic detection and classification of facial Action processing techniques called "pixel transformation" are
Units (AU) as quantified files in order to distinguish used. Using this process, images can be made to be both
between unbiased and stress/unease conditions. Then, a generic and varied. One hot encoding: According to the
model for programmed recognition of facial activ ity units significance of each textual response, a numerical weight
is proposed and created using two publicly accessible was assigned to it. "Yes" is viewed as 1, and "no" as 0 in
image datasets with comments. The findings show that Table 1. The label encoder was used to convert the
there is a clear A U relevant to pressure and that the AU's categorical data into numbers. To decode into b inary code,
potency is crucially increased during pressure, leading to a use a decoder. Despite having only one state, it does not
more expressive human face. need a decoder because it enters the next state when the bit
after the first one is high.
G. Examining physiological symptoms is a reliable
stress indicator Convert to number: As discussed above, some M L
lib raries do not take categorical variables as input. Thus, the
The signals are collected by sensors implanted in
variable is categorized into numerical variables. Below are
the human body. The necessity of manually adding
highlights can make AI calculat ions tricky. If highlights are the methods to convert a categorical (string) input to a
number:
falsely reported, accuracy suffers. Sensors attached to the
human body are used to collect these signals. The demand
for individually developed highlights can be problematic. If Label Encoder: It is used to transform non-
numerical labels into nu merical labels (or no minal
highlights are wrongly identified, accuracy suffers. Deep
categorical variab les). Nu merical labels are always between
brain networks, however, u ltimately remove h ighlights
fro m basic information at all levels of brain organization. 0 and n_classes-1.
For two tasks, the powerful neural networks processed T ABLE I. VALUES AFT ER ONE -HOT
physiological data collected fro m wrist- and chest-worn ENCODING OPERAT ION
sensors.
III. EXISTING SYSTEM S. No Binary Gray code O ne – hot
1 000 000 00000001
The contemporary system's work on an xiety 2 001 001 00000010
recognition is based on digital signal processing and 3 010 011 00000100
considers factors including skin temperature, volu me of 4 011 010 00001000
5 100 110 00010000
blood, pupil size, and the surface EM G response. However, 6 101 111 00100000
other research on this topic uses a variety of physiological 7 110 101 01000000
signals and visual cues (such as head movement and eye 8 111 100 10000000
closure) to track how a person stressed is while they are at
work. But in daily use, these measurements are awkward Logistic Regression: Like all estimation methods,
and annoying. Each stress index has a threshold value that is logistic regression is a predictive analysis, like all estimation
used to compare sensor data to in order to determine the methods. Once a binary phase is established on one or mo re
level of stress.
independent variables, this technique is used. A binary
A. Disadvantages of Existing System dependent variable used in a statistical model is referred to
 A non-stationary time performance as a "logistic regression." Logistic regression is used in
usually segregates biomedical signals regression analysis to calculate a logistic model's
used for analysis. parameters. A dependent variable with two possible values
exists in a binary logistic model, wh ich has two possible

978-1-6654-6408-6/23/$31.00 ©2023 IEEE 55

Authorized licensed use limited to: BMS Institute of Technology. Downloaded on November 23,2023 at 02:41:02 UTC from IEEE Xplore. Restrictions apply.
 Proceedings of the 7th International Conference on Computing Methodologies and Communication (ICCMC-2023)
IEEE Xplore Part Number: CFP23K25-ART; ISBN: 978-1-6654-6408-6

values and is formally represented by an indicator variab le image mapping, and stress evaluation. The third module
with the values "0" and "1". Min i-Batch Gradient Descent is converts data to binary values, which Stress level is
a subset of training that can take advantage of the speed that determined, and a remedy is determined made available for
vectorization of the code provides and modify the model's reducing stress.
parameters. Depending on the batch size, changes can be
made. As the threshold value is set using Mini Batching and V. CONCLUSION
updated by gradient, it is a reliable algorithm. The secure fault diagnosis system includes pictures
recorded by cardholders to measure emp loyee an xiety
K-Nearest Neighbor (KNN) is a classifier that can levels. The image capture is executed automatically when
be used for regression analysis. To evaluate whether a client the authenticator logs in, depending on a set time interval.
needs med ication or not, a supervised learning algorith m is Based on a certain time interval, the nu mber of images taken
used. KNN classifies the dependent variable based on how is used to calculate the user's stress level. The user's stress is
similar it is to predictor factors, wh ich are to similar estimated fro m the received slides using certain useful
instances from previously collected data. conversions and classifiers. The device will then assess the
stress levels using machine learn ing tools to gather more
Dataset: Grid views of t wo datasets: one with many good results.
previously stored properties and the other created from only
numerical input as a result of property extraction. REFERENCES
Parameters as well as stress, physical demand, performance,
and anxiety, the six p rincipal co mponents enumerated by k, [1] G. Giannakakos, D. Mancuso's, and F. Chariest, "Recognizing stress and
which relate to the dataset's only parts. Condition and anxiety from face cues in photographs," Translational Signal Processing and
interruption are impo rtant considerations. Essential Control, vol. 31, no. 1, January 2017, pp. 89-101. "Stress at Work," Journal
of Management Education, vol. 5, no. 3, pp. 50–56, 1980.
properties, namely condition (no stress, time pressure, and
[2] Nisha Raichur, Nidhi Lanikai, and Priyanka Mural,"Detection of
interruption), stress, physical demand, performance, and Pressure Employing Image Analysis and T echniques for Machine Learning
frustration, fro m the raw dataset are extracted to build a new ", vol. 9, no. 3S, July 2017.
property-extracted dataset . [3] R. Bhattacharyya with S. Basu obtained from "The Economic Times"
Dataset from the 2017 OSMI Mental Health in Tech Survey.
[4] U. S. Reddy, A. V. Thota, Workplace Stress Prediction Using Machine
Users Learning Techniques, 2018 IEEE International Conference on
Admins Computational Intelligence and Informatics Innovation (ICCIC), Madurai,
India.
[5] T. Jick and R. Payne, “Stress at work,” Journal of Management
Login Education, vol.5,no.3,pp.50-56,1980.
[6] Bakker, J., Hollander’s, L., Koscielny, R., Ctenizid, M., and Sidorov a.
Stress@ Work: Form Detecting Load to Its Knowledge, Prediction, and
Management with Individualized Coaching In: The Proceedings of the Next
ACM SIGHIT Global Conference on Health Informatics. ACM; 2012, p.
673–678.
[7] OSMI Mental Health in Tech Survey Dataset, 2017
Register Activate Users [8] Deng, Y., Wu, Z., Chu, C.H., Zhang, Q., and Hsu, D.F,. “ Combinatorial
fusion sensor feature selection and fusion for stress identification”, D.F.
World Journal of Advanced Robotic Systems 2013;10(8):306.
[9] H. Anandakumar and R. Arulmurugan, “Supervised, Unsupervised and
Upload Detected Reinforcement Learning - A Detailed Perspective”, pp.429-433. 2019.
Images Images [10] A. Haldorai, A. Ramu, and M. Suriya, “Organization Internet of Things
(IoT s): Supervised, Unsupervised, and Reinforcement Learning,”
EAI/Springer Innovations in Communication and Computing, pp. 27–53,
Stress KNN Results 2020.
[11] H. Anandakumar and R. Arulmurugan, "Artificial Intelligence and
Detection
Machine Learning for Enterprise Management," 2019 International
Conference on Smart Systems and Inventive Technology (ICSSIT ),
T irunelveli, India, 2019, pp. 1265-1269.
Live [12] A. Haldorai and U. Kandaswamy, “Supervised Machine Learning
Streams Techniques in Intelligent Network Handovers,” EAI/Springer Innovations in
Communication and Computing, pp. 135–154, 2019. doi:10.1007/978-3-
KNN 030-15416-5_7
Results [13] Gjoreski, M., Gjoreski, H., Lustrek, M., Gams, M, “Continuous stress
detection using a wrist device: in laboratory and real life”, In: Proceedings
Fig 1. Architecture Diagram for Stress Detection of the 2016 ACM International Joint Conference.
[14] H. Anandakumar and R. Arulmurugan, "Machine learning based
Multi Agent Systems in Complex Networks," 2019 Third International
Fig. 1 illustrates the stress sensing system's conference on I-SMAC (IoT in Social, Mobile, Analytics and Cloud) (I-
operation. The system has three modules, and the first one SMAC), Palladam, India, 2019, pp. 300-304.
registers the employee, sends notifications, and later offers [15] Nakashima, Y., Kim, J., Flutura, S., Seiderer, A., Andre, E, “ Stress
questionnaires. A second module deals with images. Image recognition in daily work. In: ´ International Symposium on Pervasive
capture is followed by image translation to coordinates, Computing Paradigms for Mental Health. Springer”, 2015, p. 23– 33.

978-1-6654-6408-6/23/$31.00 ©2023 IEEE 56

Authorized licensed use limited to: BMS Institute of Technology. Downloaded on November 23,2023 at 02:41:02 UTC from IEEE Xplore. Restrictions apply.