Detection and Monitoring of Work-Related Stress Using Heart Rate Variability

Stress, defined as a natural response by an organism to an intrinsic or extrinsic situation being positive, negative, physical, or mental (Unai Zalabarria et al., 2020) is undeniably one of the leading causes of physical and mental illnesses in humans (Felix Adochiei et al., 2019). It is reported to be the second most disease-causing factor in Europe and the United States. Out of four visits to the doctor, three are as a result of stress-related illnesses with some of these pathologies as serious as leading to death (Niraj K. Jha et al., 2017). A major source of stress in recent times is at the workplace, due to the intense mental and physical efforts that are required of workers and the needs of workers not being met as well as inadequate resources to promote work effectiveness and efficiency (Onur Parlak, 2021). Work-related stress has been shown to have several repercussions not only on the productivity of workers but also on the state at large. Approximately €617 billion is spent by the EU annually to cater for work-related stress depression, health costs and social welfare (Giorgia Acerbi et al., 2017). For this reason, work-related stress monitoring and management have recently become a fast-growing research field. This study aims to build on methods and algorithms for detecting and monitoring work-related stress using heart rate variability, to combat its associated adverse effects. The SWELL dataset, collected by researchers at the Institute of Computing and Information Sciences at Radboud University and the Delft University of Technology in The Netherlands was used for the implementation of the study. The dataset, consisting of a series of ECG signals were prepared for heart rate and heart rate variability feature extraction after which it was evident, significant differences between the heart rate and the heart rate variability values. The results showed an increase in the LF/LH and LF values and, a decrease in the frequency domain index HF and the time domain measurements particularly, RMSSD thus, indicating autonomic nervous system activity and hence a detection of stress. A neural network was also created using machine learning techniques on MATLAB to implement the aim aforementioned using the heart rate variability features extracted from the data set. The model was also compared with other conventional models to determine which was best for the stress detection algorithm. Overall, the goal of this study was achieved, and the chosen classification model (Artificial Neural Network) was proved to be the best for the detection and classification of stress with an accuracy and error rate of 78% and 22% respectively.