Mental Stress Level Detection using Physiological Signals and Evolutionary Algorithm

Stress can be defined as the sum of the physical, mental and emotional strains or tensions on a person. Providing a method for measuring individual's stress levels under different conditions can pave the way for early diagnosis and efficient treatments of mental related diseases. Recent studies show that some of the biological signals such as Galvanic Skin Response (GSR) and Photoplethysmogram (PPG) are associated with stress and therefore, can be used for non-invasive monitoring of stress. In this study, in addition to GSR and PPG, Abdominal Respiratory (AR) and Thoracic Respiratory (TR) signals were used as well. Various temporal and spectral domain features together with some statistical and semantic characteristics were extracted from signals of 30 volunteers who participated in a five-phase mental task. Based on data extracted, stress levels were classified into five different groups in terms of severity. In this process, an optimal feature selection procedure was performed by the NSGA-II evolutionary algorithm, and the classification step was performed by subspace discriminant classifier which resulted in 68.79% accuracy and 83.82% Interclass correlation (ICC).