Entropy-Based Emotion Recognition Using EEG Signals

Automatic detection of emotional states using electroencephalogram (EEG) signals has emerged as an attractive research topic due to advancements in brain-computer interfaces (BCIs). Considering the non-stationary nature of EEG signals, extracting emotion-related features remains a challenging task. Additionally, practical applications demand a balance between reducing computational complexity and maintaining high classification accuracy. To address these challenges, this study pioneers the application of multiscale fluctuation-based dispersion entropy (MFDE) and refined composite MFDE (RCMFDE) for emotion recognition, marking the first exploration of these features in this context. By leveraging these advanced entropy-based methods, the proposed approach significantly enhances classification accuracy while maintaining computational efficiency. The method employs optimal channel selection and data balancing strategies to provide both superior EEG characterization and reduced computational times compared to other entropy-based methods. Comprehensive evaluations on the DEAP dataset achieved binary classification accuracies of 93.51% (HA/LA) and 92.91% (HV/LV), with a multi-class accuracy of 96.67% across four classes. These results demonstrate the effectiveness of MFDE and RCMFDE in addressing critical challenges in emotion recognition and showcase their potential for real-world BCI applications. The proposed method outperforms other state-of-the-art approaches.