Stress recognition in a simulated city environment using Skin Potential Response (SPR) signals

In this paper we propose a Machine Learning (ML) classification algorithm, for stress recognition in subjects driving in a simulated city environment. Two Skin Potential Response (SPR) signals, one from each hand, are logged from the subjects and processed to remove possible motion artifacts that could appear in the recordings. This is achieved using a motion artifact removal algorithm that, taking as input the two SPR signals, outputs a single processed SPR signal. We define a number of statistical features extracted from this resulting SPR signal, and use them in a supervised learning algorithm which allows the classification of each time interval as characterized by stress or by the absence of stress. The experiments have been performed in laboratory, using a driving simulator with a motorized motion platform. The subjects had to drive, for a certain amount of time, in a city environment in two different scenarios, characterized by the presence or absence of traffic. Our findings indicate that the use of the SPR signals and of the ML classifier allow the recognition of stress situations, also showing that, in our experiment, the subjects result to be less stressed in the traffic-free scenario.