همایش ملی بیوانفورماتیک ایران

صفحه اصلی / 4th international edition and 13th Iranian Conference on Bioinformatics

Comprehensive Analysis of EEG Signals for Machine Learning-Based Depression Detection

نویسندگان :

Mikaeil Tabarraei¹ Sepideh Jabbari²

1- Department of Electrical and Computer Engineering University of Zanjan, 2- Department of Electrical and Computer Engineering University of Zanjan,

کلمات کلیدی :

Depression،Major depressive disorder (MDD)،Machine learning،Electroencephalogram (EEG)،Computer-aided diagnosis (CAD)

چکیده :

Major Depressive Disorder (MDD) is a prevalent mental illness that is typically diagnosed using questionnaire-based methods. However, these approaches often lack precision. As a result, recent research has increasingly focused on leveraging machine learning techniques for depression diagnosis. This study proposes a novel machine learning-based approach for detecting depression. The method was evaluated using electroencephalogram (EEG) data involving 59 depressed individuals and 49 healthy controls. The resting-state EEG signals were recorded in the eye-closed (EC) condition and sampled at 250 Hz. The EEG signals were recorded using a 19-channel cap, with the electrodes precisely arranged according to the internationally standardized 10-20 placement system. (Jasper, 1958). Initially, the data were preprocessed using a customized version of the Harvard Automated Processing Pipeline (HAPPE) (Gabard-Durnam et al., 2018) , adapted for EEGLAB functions (Delorme and Makeig, 2004) running on MATLAB 2020b. Artifacts caused by power-line noise, eye blinks, and muscle activity were removed during preprocessing. Subsequently, a combination of statistical, spectral, wavelet, and nonlinear features was extracted across three domains—time, frequency, and time-frequency—constructing the feature matrix. Each feature was scaled to a standardized range using the min-max normalization technique. Finally, the classification process was conducted utilizing Support Vector Machine (SVM), k-Nearest Neighbors (KNN), and Random Forest (RF) algorithms. The best accuracies achieved were (SVM = 96.45%, KNN = 91.67%, RF = 99.23%) in the time domain, (SVM = 92.16%, KNN = 93.62%, RF = 94%) in the frequency domain, and (SVM = 98.73%, KNN = 98.32%, RF = 99%) in the time-frequency domain. Additionally, the Sequential Forward Floating Selection (SFFS) method was employed for optimal feature selection, and the Sequential Backward Selection (SBS) method was applied for channel selection, resulting in accuracy improvements of 2.37% in the time domain and 6.18% in the frequency domain. These findings indicate that the proposed method is a promising approach for detecting depression using EEG signals. It has the potential to serve as a reliable supplementary tool for supporting the clinical diagnosis of depression.