Data adaptive filtering approach to improve the classification accuracy of motor imagery for BCI

This paper presents a data adaptive pre-filtering method to enhance the motor imagery (MI) classification for brain computer interface (BCI) implementation electroencephalography (EEG). The filtering is implemented using multivariate empirical mode decomposition (MEMD). The MEMD decomposes multivariate EEG signal into a finite set of time varying basis functions called intrinsic mode functions (IMFs). Several IMFs are suppressed to improve the intelligibility of the EEG for better classification. The features from pre-filtered EEG are extracted by using common spatial pattern (CSP) which is a spatial filtering approach. The two class MI classification is performed by support vector machine (SVM). The proposed method is evaluated by BCI competition III of dataset IVa. Its performance is higher compared to that of the recently developed algorithm.