Improved Non-Contact Mental Stress Detection via Bioradar

Chronic mental stress may lead to different physiological and psychological disorders, such as hypertension, depression, diabetes, obesity, cardiovascular diseases. At present there are no reliable accurate and non-subjective methods applicable for routine everyday stress assessment. In present work we proposed an improved unobtrusive machine-learning based method for remote mental stress detection by a bioradar. It was validates using experimental data of 34 practically healthy volunteers of different fitness state and stress tolerance. A CatBoost Classifier used for classify between relaxed and stress states on the volunteers allows detecting a mental stress with an accuracy of 89 %, and Fl-score of 88 %. The achieved results are comparable with ones for bioradar-based state-of-the-art techniques in terms of accuracy. Thus, the proposed method can be positioned as an easy to use unobtrusive prolonged stress detection technique suitable for people with different level of fitness and stress tolerance.