A new approach for emotions recognition through EOG and EMG signals

In this paper, an approach for emotion recognition using physiological signals has been development. The main purpose of this paper is to provide improved method for emotion recognition using horizontal electrooculogram, vertical electrooculogram, zygomaticus major electromyogram and trapezius electromyogram signals. Emotions are state of feeling that causes psychological and physical changes which affects our behaviour. Emotions are elicited using stimuli which include video, images and audio, etc. Here, emotions are elicited by audio-visual songs. For classification of emotions, time domain, frequency domain and entropy based features are extracted. These features are classified using support vector machine, naive Bayes and artificial neural network. The performance of each classifier and features is compared on the basis of accuracy, average precision and average recall. Primary contribution is the identification of time domain features as best features for EOG and EMG signals with ANN classifier to achieve maximum classification accuracy. Overall classification average accuracy (98%) of ANN is found best as compared to other classifiers. Global implications of this work is in utilization for artificial intelligence based models of human decision making systems by adding effect of emotions during decision making process modeling.