A New Approach for Enhancing Human Emotion Recognition from Electroencephalogram Signals Based on Discrete Wavelet Packet Transform

This paper presents a promising approach to enhance emotion recognition from electroencephalogram (EEG) signals by tackling the issue of outliers in the Discrete Wavelet Packet Transform (DWPT) decomposition tree.The proposed method, called Nearest-Neighbor Grubbs DWPT (NN-Grubbs-DWPT), combines nearest neighbor replacement and Grubbs' test to accurately identify and remove outliers in the DWPT decomposition tree, leading to improved quality of the extracted features and overall classification performance.The effectiveness of the NN-Grubbs-DWPT method was evaluated on two sub-datasets of the DEAP database, where the extracted features were classified using the LSSVM classifier, resulting in the classification of nine different human emotions.Our experimental results show that NN-Grubbs-DWPT outperforms the original DWPT method regarding multiple evaluation criteria, including precision, AUC, and F-score.Additionally, the NN-Grubbs-DWPT method achieved superior maximum and mean valence and arousal accuracies compared to the original DWPT algorithm.These promising results underscore the importance of addressing outliers in EEG signal processing and suggest that the NN-Grubbs-DWPT method can significantly enhance emotion recognition accuracy in practical applications of emotion recognition systems.