Filter ensemble regularized common spatial pattern for EEG classification

Abstract

Common Spatial Pattern (CSP) is one of the most effective feature extraction algorithm for Brain-Computer Interfaces (BCI). Despite its advantages of wide versatility and high efficiency, CSP is shown to be non-robust to noise and prone to over fitting when training sample number is limited. In order to overcome these problems, Regularized Common Spatial Pattern (RCSP) is further proposed. RCSP regularized covariance matrix estimation by two parameters, which reduces the estimation difference and improves the stationarity under small sample condition. However, RCSP does not make full use of the frequency information. In this paper, we presents a filter ensemble technique for RCSP (FERCSP) to further extract frequency information and aggregate all the RCSPs efficiently to get an ensemble-based solution. The performance of the proposed algorithm is evaluated on data set IVa of BCI Competition III against other five RCSPbased algorithms. The experimental results show that FERCSP significantly outperforms those of the existing methods in classification accuracy. The FERCSP outperforms the CSP algorithm and R-CSP-A algorithm in all five subjects with an average improvement of 6% in accuracy.