Abstract
Major depressive disorder (MDD) remains one of the most prevalent mental health conditions worldwide, yet its diagnosis continues to rely mainly on subjective clinical assessment. Electroencephalography (EEG) provides an objective and noninvasive means of capturing neural correlates of depression, but conventional multichannel EEG systems limit practical deployment in clinical and wearable settings. In line with a data-driven approach to depression detection using minimal brain signal sensing, this study proposes a systematic framework for MDD identification based on simplified channel- and region-wise EEG analysis. Five-minute EEG recordings were segmented into 30-second segments, and 58 features from the time, frequency, and non-linear domains were extracted from individual electrodes and anatomically defined regions. Classification was performed using a support vector machine (SVM), random forest (RF), and gradient boosting (GB) under a nested subject-wise leave-one-subject-out (LOSO) validation framework with Bayesian hyper-parameter optimization and threshold-based subject-level classification. Region-wise analysis showed that the central region achieved the highest accuracy of 93.9%, followed by the temporal region at 93.4%, the frontal region at 93.3%, the occipital region at 91.6%, and the parietal region at 89.7%. At the channel level, the central electrodes C3 and Cz achieved the highest classification accuracies of 93.8% using the RF classifier, while several frontal and temporal channels also demonstrated competitive performance. Moreover, frequency-domain features alone achieved comparable performance under the same subject-wise LOSO framework, enabling reduced model complexity. These findings demonstrate that single-channel EEG, particularly from the central region, can reliably support depression detection under a rigorous subject-independent evaluation strategy, providing a practical pathway toward wearable and accessible EEG-based mental health screening.