Survey of Deep Learning and Machine Learning Approaches for Major Depressive Disorder Detection Using EEG Data

Abstract

Major depressive disorder (MDD) is a common mental health illness in which the affected person experiences chronic sadness and loses interest in activities. Traditionally, MDD is diagnosed using clinical examinations and self-report questionnaires, both of which are subjective and susceptible to error. Recent advances in electroencephalogram (EEG) analysis have combined physiological markers with machine learning (ML) and deep learning (DL) techniques, offering objective, noninvasive alternatives for MDD detection. Despite numerous studies leveraging ML and DL for EEG-based MDD identification, there is a lack of comprehensive analysis that critically evaluates the strengths and limitations of these methods. This survey paper addresses this gap by analyzing 31 studies published between 2020 and 2024, focusing on critical aspects such as dataset size, electrode configurations, preprocessing techniques, feature engineering, and the ML/DL models employed. The paper presents a taxonomy that categorizes the methods used across various studies for each step of the EEG-based depression detection pipeline and an analysis of the critical dimensions required for detecting depression based on the comprehensive analysis of dataset, suitable EEG electrode configuration, effective preprocessing techniques, and feature engineering for different model types. With such insights, this survey aims to guide future research and improve the accuracy and reliability of EEG-based MDD detection.