Depression exists on a continuum, necessitating objective EEG-based markers for early detection. This study classified 43 subjects into three depression levels (BDI-II) using SVM and LSTM models trained on EEG data during IAPS stimulus presentation. While SVM accuracy hovered near the 33% chance level, Z-score normalization in the 100–500 ms post-stimulus window improved performance. LSTM achieved 36–39% accuracy, regardless of layer or unit adjustments. These findings suggest that specific time domains are critical for classification. Future research will optimize electrode selection and model architectures to enhance predictive accuracy for objective depression assessment.
各種検索
サポート
ヘルプ