Employing PCA and t-statistical approach for feature extraction and classification of emotion from multichannel EEG signal

Abstract

To achieve a highly efficient brain-computer interface (BCI) system regarding emotion recognition from electroencephalogram (EEG) signal, the most crucial issues are feature extractions and classifier selection. This work proposes an innovative method that hybridizes the principal component analysis (PCA) and t-statistics for feature extraction. This work contributes to successfully implement spatial PCA to reduce signal dimensionality and to select the suitable features based on the t-statistical inferences among the classes. The proposed method has been applied on the SEED dataset (SJTU Emotion EEG Dataset) that yielded significant channels and features for getting higher classification accuracy. With extracted features, four classifiers– support vector machine (SVM), artificial neural network (ANN), linear discriminant analysis (LDA), and k-nearest neighbor (kNN) method were applied to classify the emotional states. The classifiers showed slightly different classification accuracies compared to each other. ANN and SVM showed the highest classification accuracy (86.57 ± 4.08 and 85.85 ± 5.72) in case of subject dependent approach. On the other hand, the proposed method provides 84.3% and 77.1% classification accuracy with ANN and SVM, respectively in case of subject independent approach. Eventually, the proposed method and its outcomes demonstrate that this proposal is better than the several existing methods in emotion recognition. Keywords: Brain-computer interface (BCI), Emotion recognition, Electroencephalogram (EEG), SEED (SJTU Emotion EEG dataset), Support vector machine (SVM), Artificial neural network (ANN), Linear discriminant analysis (LDA), k-Nearest neighbor (kNN), Classification accuracy