Estimation Of Arousal Using Decomposed Skin Conductance Features

Electrodermal response (EDR) shows characteristic signal patterns that correspond to different emotional states. The first major step in using EDR for estimation of emotional state is the separation of various tonic and phasic components. This separation of components is more challenging when the responses overlap each other as they do when responding within shorter inter-stimulus interval. A mathematical model fitting procedure, which separates these overlapping components, is used in an experiment, where participants (n=18) were shown stimuli from the International Affective Picture System (IAPS), which varied by levels of arousal and valance. The EDR signal is collected during the experiment, and features are extracted using the mathematical model fitting procedure. These features are further used, to classify the EDR signal into high versus low arousal responses. A simple k-nearest neighbor algorithm is used to classify the features with 74% accuracy. The accuracy level obtained by a single sensor emphasizes the fact that use of specific feature extraction methods for multi-sensor applications is critical to the classification accuracy. We discuss these results in relation to adaptive system trainer design where multiple biosensors are currently being explored to assess the cognitive state of the learner. Copyright 2009 ISA. All Rights Reserved