FEATURE SELECTION IN MENTAL STRESS ANALYSIS USING MULTIPLE BIOLOGICAL SIGNALS

摘要

Stress is a response of people to face up to daily mental,rnemotional or physical challenges. Continuous monitoringrnof stress levels of a subject is of key importance to understandrnand control personal stress. In this sense, differentrnbiological signals can be used, such as, heart rate (HR), respiration,rngalvanic skin response (GSR) or electric responsernof the muscles.rnIn this paper we extract a large number of featuresrnfrom the aforementioned biological signals in order to classifyrnthe levels of stress. Once we calculate these features,rnwe use a genetic algorithm combined with a least squarernlinear discriminant (LSLD) in the aim of selecting the mostrnsuitable features, considering the error of classification.rnResults show that respiration is the most useful signal inrnthe classification of stress level and specifically, entropyrnand recurrence analysis of that signal are the most relevantrnfeatures. In the case of GSR, we observe that feet arernmore sensitive to changes of the electrodermal activity thanrnhands. With respect to EMG, it is the less adequate signalrnto classify stress level.