Pattern Analysis of Stimuli-Induced and Self-Induced Emotions Based on EEG

Up to now, various findings of brain region localization associated with different emotions have been reported. However, whether these key brain regions apply to all emotional induction methods has not been fully investigated yet. Emotions are divided into self-induced and stimuli-induced according to the induction mode, and there are few data sets about self-induced. In this paper, we developed a new dataset named MSI which includes two emotional induction methods. And we focus on identifying stability across subjects and critical brain areas' consistency in the two ways of emotion extraction. We systematically evaluate the performance of popular feature extraction and pattern classification methods with the newly developed dataset called MSI for this study. Random Forest with differential entropy features achieves the average accuracies of 81.15% in stimuli-induced emotions and 81.10% in self-induced emotions on the MSI dataset. The performance of our model shows that self-induced emotions and stimuli-induced emotions have stable recognition patterns across subjects. Further, we used the mRMR algorithm to sort each dimensional feature on the electrodes and localize the brain regions associated with emotion production. We find that the prefrontal, temporal, and occipital lobes are most associated with emotion production in both self-induced and stimulus-induced emotions. This indicates that stimuli-induced and self-induced stimuli have similarities in the physiological mechanism of emotion production.