Automated Human Emotion Recognition System Using TQWT-Based EEG Subbands

This letter presents a new framework for the identification of human emotion states, namely, positive, neutral, and negative, by using the electroencephalogram (EEG) signals. The methodology comprises advanced signal processing techniques and machine learning algorithms. The EEG signals were decomposed to various subbands by using the tunable-Q wavelet transform (TQWT). Further, from each subband, features, such as TQWT energy, total Shannon energy, Rényi entropy, Tsallis entropy, and fractal dimension, were extracted. The obtained features were combined and tested on various machine learning classifiers. The proposed method has been validated on the publicly available SJTU Emotion EEG Dataset. The accuracy obtained for human emotion recognition was 86.67% for subject-independent analysis and 88.87% for subject-dependent analysis. Also, we concluded that human emotions could be recognized more efficiently by both audio and visual stimuli as compared to individual audio or visual stimuli based on the channels selection method.