Subject-Dependent EEG-based Emotion Recognition: Preliminary Results from the EPOK Dataset

Emotion recognition constitutes an important research area for understanding individuals’ mental states and enhancing human–computer interaction.In recent years, EEG-based brain–computer interfaces enable emotions to be detected more objectively and reliably. In this study, EEG-based emotion recognition is performed using traditional machine learning methods, specifically Support Vector Machine (SVM) and Logistic Regression (LR). As part of this preliminary evaluation conducted on the EPOK EEG dataset, which is developedusing visual stimuli, emotional states are classified separately along the dimensions of arousal (low/high) and valence (low/high). In the preprocessing stage,raw EEG signals are divided into sub-frequency bands, and their Differential Entropy (DE) is calculated. Finally, the signals are classified using SVM and LRclassifiers. The performance of the models is evaluated based on accuracy, precision, recall, and F1-score. For SVM, the classification accuracies for valenceand arousal are 73.14±5.20% and 78.35±6.81% respectively. In contrast, for LR,the accuracies are calculated as 71.79±5.15% and 77.20±7.62% for valence andarousal, respectively. This study introduces the newly developed EPOK EEG dataset to the literature and demonstrates that effective emotion recognition can beachieved with visual stimuli and a low-cost EEG device.