Deep belief network based affect recognition from physiological signals

Abstract

Emotional state of a human being provides significant information to design and develop human-computer interaction based applications. Physiological signals such as Electroencephalogram (EEG), Electromyogram (EMG), Electrooculography (EOG), Skin conductance etc. play a crucial role to know the emotional state of human being during social interactions. Recently, Deep Learning has been draw attention of researchers due to significant advantages over classical approaches. In this paper, we have proposed and implemented a affect recognition system to examine emotional state of human being based on Deep Belief Network (DBN). The experiments are performed on benchmark DEAP database with two/three class of valence, arousal and dominance. We have achieved promising results with 78.28%, 70.33%, 70.16% accuracy for valence, arousal and dominance respectively.