Classification of motor imaginary EEG signals using machine learning

Abstract

Brain Computer Interface (BCI) is a term that was first introduced by Jacques Vidal in the 1970s when he created a system that can determine the human eye gaze direction, making the system able to determine the direction a person want to go or move something to using scalp-recorded visual evoked potential (VEP) over the visual cortex. Ever since that time, many researchers where captivated by the huge potential and list of possibilities that can be achieved if simply a digital machine can interpret human thoughts. In this work, we explore electroencephalography (EEG) signal classification, specifically for motor imagery (MI) tasks. Classification of MI tasks can be carried out by using machine learning and deep learning models, yet there is a trade between accuracy and computation time that needs to be maintained. The objective is to create a machine learning model that can be optimized for real-time classification while having a relatively acceptable classification accuracy. The proposed model relies on common spatial patter (CSP) for feature extraction as well as linear discriminant analysis (LDA) for classification. With simple pre-processing stage and a proper selection of data for training the model proved to have a balanced accuracy of +80% while maintaining small run-time (milliseconds) that is opted for real-time classifications.