Detection of Depressive Disorder Using EEG Signals and Machine Learning
Vignesh Rajagopalan
Monta Vista High School, Cupertino, CA, USA
Publication date: November 20, 2025
Monta Vista High School, Cupertino, CA, USA
Publication date: November 20, 2025
DOI: http://doi.org/10.34614/JIYRC2025II06
ABSTRACT
Depressive disorder, affecting approximately 280 million people worldwide, is a leading cause of disability and shortened lifespan. Traditional diagnosis relies on subjective assessments, leading to misdiagnosis and delayed treatment. The aim of this project was to develop a machine learning model for detecting depressive disorder using EEG signals, with reduced features, and identify objective biomarkers. The dataset included preprocessed EEG data from 100 depressed patients and 95 healthy controls, with 1,140 features representing Power Spectral Density and Coherence across EEG channels. Two feature engineering methods were compared: Pearson Correlation Coefficient (PCC) and Maximal Information Coefficient (MIC). PCC was calculated among features and the target, followed by iterative feature selection and transformation. MIC was calculated between each feature and the target, and the top features with the highest MIC values were directly chosen. The results from both approaches yielded the highest accuracy of 87.5%. Both methods identified alpha coherence channels as most influential on accuracy, making them potential biomarkers for depressive disorder.
Depressive disorder, affecting approximately 280 million people worldwide, is a leading cause of disability and shortened lifespan. Traditional diagnosis relies on subjective assessments, leading to misdiagnosis and delayed treatment. The aim of this project was to develop a machine learning model for detecting depressive disorder using EEG signals, with reduced features, and identify objective biomarkers. The dataset included preprocessed EEG data from 100 depressed patients and 95 healthy controls, with 1,140 features representing Power Spectral Density and Coherence across EEG channels. Two feature engineering methods were compared: Pearson Correlation Coefficient (PCC) and Maximal Information Coefficient (MIC). PCC was calculated among features and the target, followed by iterative feature selection and transformation. MIC was calculated between each feature and the target, and the top features with the highest MIC values were directly chosen. The results from both approaches yielded the highest accuracy of 87.5%. Both methods identified alpha coherence channels as most influential on accuracy, making them potential biomarkers for depressive disorder.
