Recognition of Human Mental Stress Using Machine Learning: A Case Study on Refugees

Abstract

This paper introduces a study on stress recognition utilizing mobile EEG and GSR sensors. The research involved collecting samples from a group of 55 refugees who participated in Virtual Reality stress-reduction sessions. The timing of the study coincided with an influx of refugees, prompting the development of software specifically designed to alleviate acute stress among them. The paper focuses on presenting an EEG/GSR signals pipeline for classifying stress levels, emphasizing selecting the most informative features. The classification process employed popular machine learning methods, yielding results of 86.7% for two-stress-level classification and 82.3% and 67.7% for the three- and five-level classifications, respectively. Most importantly, the positive impact of the system has been proven by subjective assessment in alignment with objective features analysis. Such a system has not yet reached the level of autonomy, but it can be a valuable support tool for mental health professionals.