Correlating substance use disorder classifications from clinical assessments and brain-computer interface experiments using unsupervised machine learning

Abstract

This work presents two investigations of clustering human behaviors resulting from substance use which are interpreted from graphs created using unsupervised machine learning. These investigations involve two datasets namely, the clinical assessments dataset and the brain-computer interface experiments dataset. The clinical assessments dataset is obtained from the National Survey on Drug Use and Health online database which is based on the Diagnostic and Statistical Manual version 5.0 (DSM-5) of mental health disorders. On the other hand, the brain-computer interface experiments dataset is obtained from an experiment of recording electroencephalogram (EEG) signals from volunteers while they consume alcoholic beverages for a specific period of time. The advantage of using unsupervised machine learning in this study is that it is capable of automatically searching for hidden patterns or groupings resulting in correlations within the dataset. These correlations are shown in the clustering graphs that visualize the relationships among human behaviors for analysis. In the clinical assessments dataset, we deduce the corresponding human behaviors from the data correlations shown in the clustering graphs. These interpretations are then validated by a mental health professional. In the brain-computer interface experiments dataset, the recorded EEG signals are analyzed using signal processing, and the corresponding human behaviors are deduced. These behaviors are then validated using the clustering graphs resulting from unsupervised machine learning.