Machine Learning Approach for Classification of Severity of Depression using Wireless EEG among Young Adults

Depression is a common mental disorder that can lead to suicide, making early treatment crucial. EEG can detect brain abnormalities associated with depression, making it a valuable tool for depression screening. This study proposes a machine learning based method for multi-level depression severity screening among university students using wireless EEG. The EMOTIV EPOC+ headset was used to record EEG signals from 30 participants (aged 18 to 25 years) with varying depression severity levels, as determined by their PHQ-9 scores. Based on these scores, the recorded EEG data was divided into five groups according to depression severity. The raw data was then processed by filtering it into sub-bands, and Hjorth parameters were extracted as features. The Ensemble Bagged Tree classifier was applied using 5-fold cross-validation. Among all the sub-bands, the beta band ($12-30 \mathrm{~Hz}$) achieved the best performance, with an accuracy of $92.52 \%$, precision of $75.08 \%$, negative predictive value (NPV) of $98.33 \%$, sensitivity of $93.72 \%$, specificity of $92.22 \%$, and an F1 score of 0.83.