On the use of autoregressive EEG modeling and support vector machine for emotional responses to musical video clips

Abstract

Brain signals convey information about emotions, even when one tries to hide them. This research proposes a method that separates emotional responses to musical video clips. Initially, autoregressive (AR) are computed for electroencephalography signals from the "Database for Emotion Analysis using Physiological signals" (DEAP). A lower bound for the AR-model order is obtained by the minimum description length criterion. Then, the AR order is chosen between this lower bound (eight) and twenty. The corresponding AR coefficients are applied as inputs to classifiers—a linear discriminator and a support vector machine. Results indicate that stress samples can be detected with an 86.3% accuracy by SVM applied to AR(8) coefficients.