Improving the separability of motor imagery EEG signals using a cross correlation-based least square support vector machine for brain-computer interface

Although brain-computer interface (BCI) techniques have been developing quickly in recent decades, there still exist a number of unsolved problems, such as improvement of motor imagery (MI) signal classification. In this paper, we propose a hybrid algorithm to improve the classification success rate of MI-based electroencephalogram (EEG) signals in BCIs. The proposed scheme develops a novel cross-correlation based feature extractor, which is aided with a least square support vector machine (LS-SVM) for two-class MI signals recognition. To verify the effectiveness of the proposed classifier, we replace the LS-SVM classifier by a logistic regression classifier and a kernel logistic regression classifier, separately, with the same features extracted from the cross-correlation technique for the classification. The proposed approach is tested on datasets, IVa and IVb of BCI Competition III. The performances of those methods are evaluated with classification accuracy through a 10-fold cross-validation procedure. We also assess the performance of the proposed method by comparing it with eight recently reported algorithms. Experimental results on the two datasets show that the proposed LS-SVM classifier provides an improvement compared to the logistic regression and kernel logistic regression classifiers. The results also indicate that the proposed approach outperforms the most recently reported eight methods and achieves a 7.40% improvement over the best results of the other eight studies.