Mental Task Classification using Machine Learning

Brain-computer interface (BCI) applications are an exciting area of research that has the potential to drastically alter the way we engage with technology. By translating brainwaves into an operational signal, it offers its users an additional means of controlling their environment. There are various techniques available to record brain activity including EEG (Electroencephalogram), fNIRS (functional near-infrared spectroscopy), and DTI (diffusion tensor imaging). Because of the low cost of acquisition devices and the lack of any known clinical risks associated with EEG, it has been widely used in the construction of brain-computer interface systems. In this work, an EEG data is recorded from 5 subjects, in three different sessions following motor imagery, motor movement, and resting state protocols. Using this dataset, we have trained different machine learning models to perform binary class classification of mental tasks into task and non-task related ones. For taskrelated data, we have used Motor Imagery (MI) data while the non-task related involves the Resting state (eyes open) recording of data. We have evaluated the performances of four machine learning algorithms: Decision Tree, KNN, Ensembles, and SVM, out of which SVM performed the best with the training and test accuracy of 87.9 and 72.7% respectively.