Some considerations of common spatial pattern for better classification in brain-computer interfaces

Abstract

EEG-based motor imagery signal classification is very important in brain-computer interface (BCI) technology. In this work, we develop a common spatial pattern (CSP) technique for feature extraction in a BCI system. To confirm classification improvement, classification accuracy was analysed by using four statistics, namely mean, variance, skewness, and kurtosis within the CSP paradigm. The data from the dataset III of BCI competition II were used and simulated using MATLAB. The results show that the best classification accuracy is obtained when the CSP algorithm uses the variance statistic for feature extraction.