Detection of Stress Levels Using Biomedical Signals and Artificial Intelligence

Stress is a state that occurs when an individual's physical and mental resources are taxed in response to demands, becoming especially evident under heavy mental exertion. Mental workload is a significant psychophysiological metric that directly influences task performance and can also lead to mental diseases such as depression. Thus, the objective evaluation of stress levels using physiological data is crucial for enhancing work productivity and assuring safety. This work employed an integrated approach utilizing electrocardiography (ECG) and photoplethysmography (PPG) signals for stress detection. The data were sourced from the publically accessible MAUS dataset and gathered from 22 healthy participants utilizing wearable sensors during N-back activities. The signals were segmented into epochs, and a total of 50 features were extracted at both temporal and spectral levels. The features were examined utilizing diverse machine learning algorithms. The models' performance is assessed using accuracy, specificity, F-score, and AUC criteria, with the Bagged Trees method achieving the greatest accuracy of 98.6%. The results indicate that employing several biosignals and sophisticated signal processing techniques provides excellent precision in stress detection. The device provides a pragmatic option for real-time monitoring of individuals' stress levels in their daily lives, thanks to its portable design.