Analyzing Electrodermal Activity Data With An Unsupervised Machine Learning Approach
Electrodermal activity (EDA) is a psychophysiological index that is used as a measure of arousal in the sympathetic nervous system and thus as an indicator of whether an individual is eliciting a physiological response to stress. Density Based Spatial Clustering of Applications with Noise (DBSCAN) and k means were used to compare the structure of clusters formed from EDA signals from forty-three participants acquired from Biopac and the Empatica E4 during a psychological study designed to elicit a stress response. Window lengths 3, 5, 10, 30, 60, 90 and 120 were used to calculate input features. A novel visualization approach to analyzing cluster assignments constructed from time-series data is presented. Results suggest that a higher window length should be used for signal-based analyses and that data collected from the Empatica E4 are able to identify regions of stress at a level comparable to signals collected from Biopac.