TY - GEN

T1 - Depression Detection Through EEG Signal Analysis: A Convolutional Autoencoder Deep Learning Model

AU - Firoz, Neda

AU - Aksyonov, Sergey Vladimirovich

AU - Berestneva, Olga Grigorievna

AU - Savostyanov, Alexander

N1 - Conference code: 26

PY - 2025/8/8

Y1 - 2025/8/8

N2 - Depression is a debilitating and enervating mental health disorder that requires attention for necessitating accurate and efficient diagnostic techniques. Developments in deep learning and neurophysiological data analysis have enabled the use of EEG signals for binary depression classification. In this study, a novel approach is introduced that utilizes Convolutional Autoencoders for feature extraction from EEG signals, to enhance feature representation for classification of depression. To our knowledge, this is the first study utilizing the ICBrainDB dataset, which includes EEG test results and psychological questionnaire responses from over 1,000 participants across various regions of Russia. A series of experiments were carried out to evaluate the classification performance of both traditional machine learning and deep learning approaches in predicting depression using this novel dataset. The findings demonstrate that incorporating EEG feature sets extracted through CAE encodings significantly enhances classification accuracy. Specifically, the Random Forest and CNN models achieved impressive classification accuracies of 98.31% and 99.31%, respectively, in distinguishing individuals with depression from healthy controls. This study contributes to the expanding field of computational psychiatry by introducing a robust, data-driven framework for depression prediction, fostering the development of more reliable and automated mental health assessments.

AB - Depression is a debilitating and enervating mental health disorder that requires attention for necessitating accurate and efficient diagnostic techniques. Developments in deep learning and neurophysiological data analysis have enabled the use of EEG signals for binary depression classification. In this study, a novel approach is introduced that utilizes Convolutional Autoencoders for feature extraction from EEG signals, to enhance feature representation for classification of depression. To our knowledge, this is the first study utilizing the ICBrainDB dataset, which includes EEG test results and psychological questionnaire responses from over 1,000 participants across various regions of Russia. A series of experiments were carried out to evaluate the classification performance of both traditional machine learning and deep learning approaches in predicting depression using this novel dataset. The findings demonstrate that incorporating EEG feature sets extracted through CAE encodings significantly enhances classification accuracy. Specifically, the Random Forest and CNN models achieved impressive classification accuracies of 98.31% and 99.31%, respectively, in distinguishing individuals with depression from healthy controls. This study contributes to the expanding field of computational psychiatry by introducing a robust, data-driven framework for depression prediction, fostering the development of more reliable and automated mental health assessments.

KW - Convolutional Autoencoders

KW - Deep learning

KW - Depression

KW - EEG signals

KW - ICBrainDB

KW - Machine learning

UR - https://www.scopus.com/pages/publications/105014193983

UR - https://www.mendeley.com/catalogue/0725e6e9-7ea0-3c31-90ef-b7974486417e/

U2 - 10.1109/EDM65517.2025.11096865

DO - 10.1109/EDM65517.2025.11096865

M3 - Conference contribution

SN - 9781665477376

T3 - International Conference of Young Specialists on Micro/Nanotechnologies and Electron Devices, EDM

SP - 1820

EP - 1825

BT - International Conference of Young Specialists on Micro/Nanotechnologies and Electron Devices, EDM

PB - IEEE Computer Society

T2 - 2025 IEEE 26th International Conference of Young Professionals in Electron Devices and Materials (EDM)

Y2 - 27 June 2025 through 1 July 2025

ER -