Optimizing Eeg Channel Selection by Regularized Spatial Filtering and Multi Band Signal Decomposition

Appropriate choice of number of electrodes and their positions are essential in Brain-Computer Interface applications since using less electrodes collects insufficient information for classification purposes whereas using more collects redundant information that could degrade BCI performance. This paper proposes a novel method of optimizing EEG channel selection by using the regularized Common Spatial Pattern (CSP) algorithm to discard redundant channels and multi band signal decomposition to select subject-specific frequency range. The performance of the proposed method is compared with EEG channel selection using Fisher criterion, mutual information, support vector and CSP on 9 subjects for two motor imagery tasks. Experiment results show the proposed method yields the highest accuracy in selecting 4 to 10 channels compared with the methods studied as well as using all the channels. The results also illustrate the proposed method significantly improves by multi band filtering and can achieve an average of 47% reduction of channels with only an averaged drop of 1.04% in classification accuracy.