Detection of Music-Induced Emotion Changes by Functional Brain Networks

Abstract

This chapter discusses emotions induced by music and attempts to detect emotional states based on regional interactions within the brain. The brain network theory largely attributes statistical measures of interdependence as indicators of brain region interactions/connectivity. In this paper, the authors studied two bivariate models of brain connectivity and employed thresholding based on relative values among electrode pairs, in order to give a multivariate flavor to these models. The experimental results suggest that thresholding the brain connectivity measures based on their relative strength increase classification accuracy by approximately 10% and 8% in time domain and frequency domain respectively. The results are based on emotion recognition accuracy obtained by decision tree based linear support vector machines, considering the thresholded connectivity measures as features. The emotions were categorized as fear, happiness, sadness, and relaxation.