PhD thesis defense to be held on June 14, 2023, at 12:30 (Multimedia/Tele-teaching Room, NTUA Central Library)
Picture Credit: Thelma Androutsou
Thesis title: Development of an Automated Occupational Stress Detection System based on Physiological and Behavioural Parameters Extracted from Data Analysis during Computer Use
Abstract: Occupational stress is an open and major challenge for modern societies, as the rapid development of technology and new forms of the global economy have brought about significant changes in the nature of the workplace and the conditions prevailing in it. The field of Affective Computing includes a large variety of studies and tools that, by employing the analysis of psychological, physiological, and behavioral parameters, can lead to automatic stress detection and monitoring and, consequently, effective stress management. As regards the workplace sector, in recent years the need has emerged to develop unobtrusive systems that do not disturb the work pattern and physical behavior of users. In this direction, there is a growing research interest in harnessing devices used daily in the workplace in order to monitor the well-being and performance of workers, through the use of sensors and innovative data analysis methods. The personal computer and its peripherals are among the prime candidates for the implementation of such systems. The aim of this thesis is the design and development of a non-invasive system that analyzes the data from the computer use in office work and automatically monitors and detects occupational stress through physiological and behavioral parameters extracted from them. In this context, a smart device was constructed and programmed, the structure of which consists of physiological signal recording sensors and a development board including a microcontroller and a Wi-Fi module, integrated in a commercially available, wired computer mouse. Furthermore, an experimental protocol was designed and implemented to simulate an office environment and include the most known occupational stressors. The collected data, which included physiological measurements from system sensors, behavioral measurements from the use of the computer keyboard and mouse, and psychological parameters from questionnaires, were subjected to processing and filtering methods to isolate useful information and extract parameters. Statistical analysis tools were applied to validate both the effectiveness of the experimental protocol used and the reliability of the proposed device in recording and monitoring the physiological signals of users. Additionally, machine learning models were trained using known classifiers and exploring different data annotation methods. Particularly high-performance metrics were observed in models trained individually with physiological and behavioral parameters, while the feature-level fusion analysis successfully detected stress with an accuracy exceeding 90% and F1 score of 0.90. The decision-level fusion analysis, combining the features extracted from both the computer mouse and keyboard, showed an average accuracy of 66% and an average F1 score of 0.56.
Supervisor: Professor Emeritus Dimitrios Koutsouris
PhD Student: Thelma Androutsou