Time-frequency decomposition-based weighted ensemble learning for motor imagery EEG classification

Abstract

Motor imagery brain-computer interface system based on Electroencephalogram (EEG) is an effective way to help the disabled recover part of their motor abilities. However, decoding the movement intention contained in the EEG signal accurately presents many challenges. In this paper, we propose a time-frequency decomposition-based weighted ensemble learning (TFDWEL) method, which aims to improve the classification performance of motor imagery EEG signals. The TFDWEL method divides the EEG signal into multiple subsets, and uses four time-frequency processing methods to extract the time-frequency sub-bands of each subset. Then the feature extraction model and classifier model of each subset trained by the common spatial pattern (CSP) algorithm and the support vector machine (SVM) algorithm are used to build a set of base learners. The least square error estimation method is used to learn the weight of each base learner, and finally the weighted summation method is used to obtain the final decision. The classification performance of the TFDWEL method is evaluated on the BCI Competition IV Data Set 2b, and the results show that the classification accuracy of 81.58% can be obtained. Superior classification performance indicates that the TFDWEL method can be used in further research to help the rehabilitation of the disabled.