Channel Selection Using Stochastic Diffusion Search Algorithm for Classification in Brain-Computer Interface

Utilization of the Brain-Computer Interfaces (BCI) is done via Electroencephalogram (EEG) signals that provide several environmental interactions among individuals having restricted movements owing to neurodegenerative diseases or strokes. However, the BCI system was based on Motor Imagery (MI). It was not used for any form of real-life application owing to a decrease in the performance of various Common Spatial Pattern (CSP) algorithms, especially while the actual number of channels was high. A multi-channel structure of such EEG signals can increase cost and bring down speed. Due to this, a reduction in the system cost by the detection of active electrodes during the process can increase accessibility. This way, optimization techniques in choosing electrodes can be used to determine other effective channels by employing a method of random selection. For this work, a Stochastic Diffusion Search (SDS) algorithm based on herd optimization techniques was used with four different classifiers, which were the AdaBoost, the Classification and Regression Tree (CART), the Naive Bayes (NB) as well as the K-Nearest Neighbor (KNN). The channels that were chosen frequently were determined to improve the system performance with regard to accuracy and speed. The results proved that the approach proposed was successful in bringing down the channel number and run time without affecting the accuracy of classification.