A comparative analysis of emotion recognition from EEG signals using temporal features and hyperparameter-tuned machine learning techniques.

Classifying emotions based on EEG signals is really important for enhancing our interactions with computers, monitoring mental health and creating applications in affective computing field. This study explores improving emotion recognition performance by applying traditional machine learning classifiers and boosting techniques to EEG data from the DEAP dataset. To categorize emotional states, we used four classifiers: K-Nearest Neighbors (KNN), Support Vector Machine (SVM), XGBoost and Gradient Boosting. Differential entropy and Higuchi's fractal dimension are two important time-domain parameters that we extracted after applying a segmentation technique to capture the temporal interdependence of EEG data. These features were selected for their ability to reflect intricate neural dynamics associated with emotional processing. A five-fold cross-validation procedure was applied to estimate the model's performance and hyperparameter tuning was conducted to optimize classifier efficiency. XGBoost achieved the highest accuracy 89 % for valence and 88 % for arousal demonstrating its superior performance. Furthermore, cross-subject evaluation on the SEED dataset reinforced the approach's robustness, where XGBoost achieved 86 % accuracy using HFD and 84 % using DE. These results emphasize the effectiveness of combining advanced feature extraction methods with boosting algorithms for EEG-based emotion recognition, offering promising directions for the development of real-world emotion-aware systems. The key findings of this research are as follows:•Differential Entropy and Higuchi's Fractal Dimension proved effective in capturing emotional brain dynamics•XGBoost outperformed other classifiers in both DEAP and SEED datasets•The proposed method demonstrates robustness across subject variations and datasets.