Recognizing Mental States when Diagnosing Psychiatric Patients via BCI and Machine Learning

Abstract

A psychiatric disorder is any disorder that interferes with a person’s thoughts, emotions, or behavior, including anxiety disorders, and depressive disorders. In order to diagnose these disorders, patients are often required to go through a series of diagnostic tests that demand concentration for accurate results. This paper aims to lower misdiagnoses that result from the lack of concentration in patients by developing an algorithm that recognizes mental states using 989 columns of EEG brain wave data. These data columns were entered into a train split function and analyzed using different classification models of Decision Tree, Logistic Regression, Random Forest, Gradient Boosting, Adaptive Boosting, K-neighbors, LGBM, and XGB. The control experiment analyzed the raw dataset, the second and third experiments utilized feature extraction algorithms of PCA and ICA analysis respectively, and the fourth experiment used correlation matrix analysis to produce accuracy scores. The highest accuracy score of 98.19% was produced by the LGBM Classifier in the control experiment and the most efficient feature selection method was PCA. The highest PCA processed data yielded an accuracy score of 87.7% using the random forest classification model while the correlation matrix analysis yielded an accuracy score of 80.04% using the LGBM classifier. These findings will allow psychologists and psychiatrists to provide methods to help patients answer all questions at a sustained level of sufficient concentration, ultimately allowing a higher percentage of proper diagnoses to be made.