Passive EEG-based emotional state cognition using subject-independent feature extraction

Abstract

Research on emotion recognition technologies, whereby emotional state is read off via using voice, facial image, bio-signal and electroencephalogram (EEG), has received considerable attention due to its prevalence in various applications including entertainment, health care, on-line education, automobile, marketing research, to robot and etc. Among the emotional state cognitive technologies utilizing the biometric data, the study on EEG is one of the most widely used. However, EEG-based emotion recognition meets challenges because of multi-channels time series data structure in which contains abnormal characteristics changing the statistical component over time. Consequently, this leads to investigate towards appropriate feature extraction approach in order to be more versatile data analysis to capture underlying information for cognitive the emotional state from EEG signals. In this paper, a subject-independent multiclass-common spatial patterns (Si-MCSP) based on passive EEG therefore is proposed to cognitive the emotional states. Si-MCSP can extract the prominent features in the terms of the subject-independence with respect to the each emotional state. Furthermore, Si-MCSP can automatically determine the numbers of the optimal features which provide an improved classification system of emotional state dramatically. Multiclass Support Vector Machine (Multi-SVM) model then is set up to classify these features for emotion recognition. The experimental results show that the proposed feature extraction method, Si-MCSP, performs higher classification accuracy than the popular feature extraction methods.