Classification of Imaginary motor task from Electroencephalographic Signals: A Comparison of Feature Selection Methods and Classification Algorithms

Abstract

In this work, a Brain Computer interface able to decode imagery motor task from EEG is presented. The method uses time-frequency representation of the brain signal recorded in different regions of the brain to extract important features. Principal Component Analysis and Sequential Forward Selection methods are compared in their ability to represent the feature set in a compact form, removing at the same time unnecessary information. Finally, two method based on machine learning are implemented for the task of classification. Results show that it is possible to decode the mental activity of the subjects with accuracy above 80%. Furthermore, visualization of the main components extracted from the brain signal allow for physiological insights on the activity that take place in the sensorimotor cortex during execution of imaginary movement of different parts of the body.