A Hybrid Deep Learning Model (CNN-GRU) with MRMR for EEG-Based Depression Detection

Background/Objectives: Major depressive disorder (MDD) is a prevalent mental health disorder for which objective and reliable diagnostic approaches remain an active area of research. Electroencephalography (EEG) has emerged as a promising non-invasive modality for identifying neural activity patterns associated with depression, while advances in deep learning have enabled effective analysis of complex EEG signals. Methods: In this study, we propose a hybrid deep learning framework that integrates convolutional neural networks (CNNs) for spatial feature extraction and gated recurrent units (GRUs) for temporal modeling of EEG signals. The spatial and temporal features extracted by the CNN and GRU components were concatenated into a 120-dimensional representation, after which minimum redundancy maximum relevance (mRMR) feature selection was applied to reduce redundancy and retain the 30 most informative features before classification. The proposed framework was evaluated using the publicly available MODMA EEG dataset, and its performance was assessed using accuracy, precision, recall, and F1-score. Results: Across 30 independent simulation runs, the CNN–GRU–mRMR framework achieved a mean classification accuracy of 96.74%, exceeding the performance of the comparison methods evaluated in this study. The model also demonstrated high precision, recall, and F1-score, indicating strong classification performance within the evaluated dataset. Conclusions: These findings suggest that integrating spatial and temporal feature learning with mRMR-based feature selection can improve EEG-based depression classification. Further validation using larger, more diverse, and independent clinical datasets is needed to establish the generalizability of the framework and evaluate its potential utility for objective EEG-based depression assessment.