An Optimized Deep Learning Based Framework for Cognitive State Classification Using Brain-Computer Interface Technologies

The most widely recognized and bitter of all mental disorders has been known to be depression, which in fact transcends all
the age cohorts hopelessly into the realm of human suffering. Such psychological discomfort resulting from delays in
recognizing depressive symptoms or delaying intervention may lead to serious reactions which may develop into suicidal
behaviors. Nevertheless, despite the wide prevalence of this condition, it remains poorly diagnosed and almost untreated owing
to various forms of persistent societal stigma as well as a lack of acknowledgment of this condition as a bona fide mental
illness. This research examined the effectiveness of six discrete machine learning classifiers for the identification of depression,
which used a combination of socio-demographic and psychological features. The features from the dataset that can serve as
significant predictors were selected using three feature selection algorithms- SelectKBest, Minimum Redundancy Maximum

Relevance (mRMR), and Boruta-for the enhancement of the performance of all models. Finally, the Synthetic Minority Over-
sampling Technique (SMOTE) was harnessed to tackle the problem of class imbalance in training data, thus improving models'

generalizability and accuracy. Among the compared models, AdaBoost classifier with SelectKBest feature selection showed the highest accuracy score value of 92.56%. Further, complete evaluation metrics such as sensitivity, specificity, precision, and F1 score along with area under the Receiver Operating Characteristic curve (AUC) were computed to give a thorough insight into the diagnostic capacity of each model. These findings further endorse the premise that such methods would have varying promise in the early and precise detection of depression among different populations.