An Enhanced Approach in Predicting and Classifying Major Depressive Disorder Using Bi-GRU with Attention Mechanism and Transfer Learning

Major Depressive Disorder (MDD) is a serious issue in medical research due to its high prevalence and significant impact on the quality of life of individuals, leading to disability, comorbidity, and an increased risk of suicide. Accurate and early diagnosis is crucial for effective treatment, yet predicting MDD remains challenging due to its complex etiology, overlapping symptoms with other psychiatric disorders, and the subjective nature of traditional diagnostic methods. This study proposes RAG-EEGNet, a novel approach to detect MDD using EEG data, integrating advanced feature extraction and selection techniques with deep learning models. Initially, Boruta was used for comprehensive feature extraction, followed by Cuckoo Search Optimization (CSO) to select impactful features. A hybrid model is employed for classification, combining ResNet-50 and a Bi-GRU enhanced by an attention mechanism. The results show a significant improvement in the detection of MDD with an accuracy of 99.01%, precision of 100%, recall of 99.24%, F1-score of 99.12%, and ROC-AUC of 99.0%, demonstrating the efficacy of the proposed approach, highlighting the critical role of EEG data in diagnosing and predicting mental diseases.