Investigation on the effect of feature extraction block on the performance of EEG-based emotion recognition system

Emotion recognition from brain signals contributes significantly to various fields, including healthcare, education, entertainment, facilitating emotion regulation and therapeutic interventions. However, existing methods face challenges due to the non-linearity of brain activity, inter/intra-variability, and artifacts. In order to improve emotion recognition from electroencephalogram (EEG) signals, this study addresses aspects related to feature extraction methods. to investigate the effect of feature extraction on the performance of EEG-based emotion recognition, using three feature extraction methods, Hjorth parameters (HP), Higuchi fractal dimension (HFD), and high-order crossing (HOC). The proposed approach is optimized by exploring optimal parameters for HFD and HOC. Extracted features were classified using linear discriminant analysis (LDA) with mutual information for feature selection. The performance of our approach is assessed in the DEAP dataset. The obtained results show that hjorth-base approach outperforms other approaches with an accuracy of 70.48% for the arousal class, and an accuracy of 68.75% for the valence class. demonstrating it’s reliability for EEG-based emotion recognition. HFD and HOC followed, underlining the importance of parameter optimization and the impact of using all 32 EEG channels for emotion classification. The comparison of the proposed approach with some state-of-the-art methods shows the efficacy and superiority of our approach.