A Novel Approach for EEG Electrode Selection in Automated Emotion Recognition Based on Lagged Poincare’s Indices and sLORETA

Abstract

The goal of this paper was to develop a novel method to track emotional processing in different brain regions using electroencephalogram (EEG) analysis. In addition, the role of EEG electrode selection and feature reduction in emotion recognition was investigated. To this end, the multi-channel EEG signals of 32 subjects available in DEAP dataset were studied. The best EEG electrode positions were selected based on lagged Poincare’s measures of EEG recordings and a source localization method (sLORETA). Three feature reduction algorithms, including random subset feature selection (RSFS), sequential floating forward selection (SFFS), and sequential forward selection (SFS) in combination with support vector machine (SVM), were evaluated to classify high/low valence and high/low arousal. The results showed that RSFS outperformed the other feature selection approaches. In addition, the positive impact of the EEG electrode selection on the classification performances has been confirmed. The most active EEG electrodes were FP1, C3, Cp1, P3, and Pz. Adopting RSFS and selected EEG electrodes, the mean subject-independent accuracies of 73.89 and 74.62% and subject-dependent accuracies of 98.97 and 98.94% were obtained for valence and arousal dimensions, respectively.