EMOTION RECOGNITION FROM EEG SIGNALS BASED ON OPTIMIZATION OF MUTUAL INFORMATION

Emotion is a mental state of feelings triggered by internal or external stimuli and sometimes by both. In general Mutual Information (MI) between the subjects' EEG signal energy and a manual score of emotion (MSE) is less which makes the prediction task difficult. In this work, to enhance the accuracy, nonlinear prediction methods are applied to normalized data sets with two classes of emotion where an emotional template has been designed based on adaptive bins for the classification of emotion. In this work, Subject-dependent emotion prediction has been explored extensively. We found that the MI or dependence between the data set and the MSE is the key to higher prediction accuracies. So it is important and necessary to find dimensions in signal space along which the MI between EEG signal energy and MSE is optimum. In particular, a gradient ascent algorithm has been used to find such dimensions in the multi-channel EEG signal energy space. In this current research, we have achieved high prediction accuracies of 86.95% to 90.23% for two classes of emotion in subject-dependent classification. The merit of the proposed approach is that it is devoid of any assumptions and user-based choice of parameters. we propose a simple but efficient emotion recognition approach using DEAP database.