Channel selection for brain signal classification by penalized automatic relevance determination

Abstract

Channel selection or reduction in Brain computer interface (BCI) is important to reduce the cost and improve the generalized accuracy. A channel selection method using group automatic relevance determination (GARD) for P300 based BCI has been reported. In this paper, we apply the penalized ARD (PARD) which is an extension of ARD, and compare with GARD in our auditory BCI. Experimental results show that PARD provides more sparse solution than GARD while PARD shows almost the same classification accuracy as GARD.