Common Spatial Pattern and Particle Swarm Optimization for Channel Selection in BCI

Common spatial pattern algorithm (CSP) is famous for extracting ERD/ERS feature from multi-channel BCIs based on motor imagery. However, if channel number is large, CSP will tend to overfitting and it is inconvenient for clinical operation. In this study, CSP filters' discrimination and channel number are integrated under one roof. Then binary particle swarm optimization (BPSO) is employed to select the best channel groups. Experimental results of BCI2003 dataset IV and BCI2005 dataset I show that good classification accuracies can be achieved only with 914 channels.