Optimization of the Channels Selection and Classification Accuracy Based on Maximum Mean Discrepancy in BCI System

In electroencephalogram (EEG) brain-computer interfaces (BCI), multichannel EEG is generally used to improve the classification accuracy. However, the performance of systems degrades when the number of channels is large. Therefore, how to select the suitable channels for classification of different motor imagery tasks by removing irrelevant or noisy channels becomes important. The optimal combination of channels selection method is presented in this paper. Based on the bias of the value of maximum mean discrepancy (MMD), every channel's contribution score is obtained under right hand and right foot motor imagery tasks from BCI competition III. Then, a small number of channels are selected by the larger MMD scores. At last, the optimal combination of channels is resorted from them by looking for a threshold p, which is calculated by the highest classification accuracy rate based on the support vector machines (SVM). The results show that this method can not only reduce the number of channels with the classification accuracy in an acceptable range, and increase the rate of convergence with very few leads. The proposed p -SVM-MMD algorithm has a better performance in classification accuracy compared to the use of the fixed channels (C3, C4, and Cz) and multichannel.