Average Power Based EEG Channel Selection Method for Emotion Recognition

Abstract

AbstractEmotion can be defined as the neurophysiological changes people experience due to significant internal or external occasions. This work applies a channel selection algorithm based on average band power on the preprocessed EEG data frequency bands from the DEAP dataset to select the top 10 EEG channels. DWT is performed to get detail coefficients as features, and a statistical parameter RSS is used to reduce the dimension of the features for both selected and all 32 EEG channels. Finally, valence and arousal are classified using different classification algorithms (like Random Forest, Extra Trees, Naive Bayes, and MLP) to make a performance comparison between selected EEG channels and all EEG channels. The highest test accuracy, 66.8%, was retrieved from the Random Forest (RF) classifier for valence classification. Likewise, Random Forest (RF) and Extra Trees (ET) both achieved the highest test accuracy of 64.84% for arousal classification, which validates the efficiency of the proposed channel reduction technique.KeywordsEmotion recognitionDEAPEEGChannel selectionDWTRSS