부산대학교 도서관

Brain-computer interface (BCI) system-based emotion classification is an intriguing topic for research. Feature extraction is an essential step in these systems, usually carried out by a deep neural network or hand-crafted features. Studies in this area frequently highlight the significance of a small set of features; a precise comparison of different proposed features is still necessary. To classify emotions from EEG data, we attempt an exploratory evaluation of a number of EEG features, including those in the time domain, frequency domain, auto-encoded, and deep features, using the DENS dataset. Baseline techniques like Random Forest (RF), XgBoost (XGB), and KNN are used to compare feature performance. The results of this study shed light on the possibility of using these EEG features to identify emotional states. Compared to the previous results on the same dataset, our results show accuracies of 82.10% and 81.33% for valence and arousal, an increase of 29.63% and 18.82%, respectively.