Mindchrono: A Comparative Study of Deep Learning Models for Eeg-based Depression Detection

Clinical depression is a widespread mental health disorder that significantly impacts an individual’s quality of life and societal productivity. Traditional diagnostic approaches primarily rely on subjective assessments, including selfreported symptoms and clinical interviews, which can lead to inconsistencies and delayed interventions. To address these challenges, this study explores an EEG-based deep learning framework that automates depression detection, ensuring objective, scalable, and real-time assessment. This study utilizes the EEG Psychiatric Disorder Dataset, which includes EEG recordings from X depression patients and Y healthy controls, ensuring a well-balanced dataset for model evaluation.. The dataset contains 1149 EEG features, including Power Spectral Density (PSD), Hjorth parameters, coherence, and quantitative EEG (qEEG) variables, capturing brainwave activity across multiple frequency bands. EEG signals undergo a comprehensive preprocessing pipeline, incorporating band-pass filtering, independent component analysis (ICA), and normalization to enhance signal clarity and reduce noise artifacts. Three state-of-theart deep learning models-Transformers, Recurrent Neural Networks (RNNs), and Long Short-Term Memory Variational Autoencoders (LSTM-VAEs)-are employed to analyze the processed EEG signals. Transformers, leveraging self-attention mechanisms, achieve the highest classification accuracy of $95 \%$, outperforming RNNs (91%) and LSTMVAEs ($89.95 \%$). The superior performance of Transformers is due to their ability to capture long-range dependencies in EEG sequences, improving feature representation and classification accuracy. This study underscores the critical role of model selection in EEG-based depression detection, demonstrating that deep learning techniques can significantly improve diagnostic reliability, automation, and scalability. By reducing reliance on subjective clinical evaluations, the proposed approach paves the way for real-time, data-driven mental health diagnostics, improving accessibility and early intervention strategies.