Classification of emotions using EEG activity associated with different areas of the brain

Abstract

• For emotion classification , emotional brain activity should be analyzed for each brain lobe and hemisphere. • The four lobes of the brain give different emotion classification accuracies as they generate distinct emotional activity. • The hemispheric regions of brain give different classification accuracies as they generate distinct emotional activity. • Fusion of EEG and peripheral signals results in better emotion classification accuracy than using EEG signals alone. It is important that interactive Affective Brain-Computer Interface (aBCI) applications support some degree of emotional intelligence. Many of the previous works that have explored the use of the Electroencephalogram (EEG) for the purpose of emotion recognition have focused on using data coming from the entire brain. However, the emotional activity in humans is not constant across the brain but varies from one region to another. Therefore, this paper aims to classify human emotions into high/low arousal and high/low valence using EEG and other physiological data from three datasets - DEAP, DREAMER and DASPS, showing the differences in the classification accuracies for different regions of the brain such as frontal lobe, parietal lobe , temporal lobe, occipital lobe, left frontal region, right frontal region, left parietal-temporal-occipital region and the right parietal-temporal-occipital region. The classification experiments are performed using the 1D Convolutional LSTM network and its performance is then compared with two baseline machine learning (ML) algorithms K-Nearest Neighbor and Random Forest. The experimental results show the variation of classification accuracies from one brain region to another. Moreover, they also indicate that the fusion of EEG and peripheral data produces higher emotion classification accuracy as compared to using the EEG and peripheral data separately.