Application of NSGA-II in Channel Selection of Motor Imagery EEG Signals with Common Spatio-Spectral Patterns in BCI Systems

Abstract

Nowadays, brain activities are the focus of research in conjunction with Brain-Computer Interface devices that are employed in broad number of products. Electroencephalography (EEG) signals have been the most frequently used method for collecting data mostly on brain's electrical activity. The common spatio-spectral pattern (CSSP) has been developed to extract features from EEG data associated with Motor Imagery (MI) tasks. CSSP is an effective method for discriminating and classifying EEG signals in two-class motor imagery signals based on movement tasks. It has been implemented and compared to conventional CSP, as well as the optimized methods for such application. The meta-heuristic multi-objective non-dominated sorting GA algorithm (NSGA-II) was used to choose the optimal subset of channels among multi-channel EEG motor imagery signals in this investigation. The objective is to develop an optimal solution to the channel selection problem in order to select the ideal subset of channels from multi-channel electroencephalogram (EEG) signals in brain-computer interface systems. The primary objective of the channel selection technique in EEG signal analysis is to obtain a limited number of channels so that the subject feels more at ease with the gel-based EEG electrodes. Additionally, this technique reduces the problem of overfitting in classification results when a significant number of redundant channels are available. Finally, the findings reveal that the proposed CSSP approach with 10-fold cross-validation achieves a higher classification accuracy than the CSP, CSP-TSM, and LRCSP methods.