Comparison of machine learning methods for two class motor imagery tasks using EEG in brain-computer interface

The Brain-Computer Interface (BCI) systems can improve the life quality of physically impaired people. It allows them to perform tasks like gripping objects, turning on light, changing the television channels, etc. In fact, the BCI is a mechanism of identifying the cerebral commands and transforming them into actions via the processor. This paper deals with the design of an effective BCI system in which EEG signals are used as brain commands. Different types of brain activities can cause EEG signals to vary, affecting the classification performance. In this study, the signal is enhanced by employing the Multiscale Principle Component Analysis (MSPCA). Features from the enhanced signal are extracted by using the Wavelet Packet Decomposition (WPD). The extracted features are employed to study various classifiers' effectiveness in the classification of the EEG signals, recorded from five different subjects while intending to move their right foot and hand. The total classification accuracy is employed for comparing the obtained results. It is shown that an effective grouping of MSPCA, WPD and Random Forest classifier achieves a total classification accuracy of 98.45%. Results conclude that the suggested approach is a potential candidate for the design and development of future BCI systems.