An EEG Based Framework For Classifying Motor Imagery Signal

Classification of motor imagery signal is one of the major part of a Brain Computer Interface (BCI) system, which possess immense potential to ease the life of physically disabled people. In this article, a generalized framework has been presented for motor imagery signal classification, with emphasis on the feature selection. A Manhattan distance based feature selection algorithm is proposed, which is solved by a greedy approach. Experimental results illustrate that the soft margin SVM classifier produces highest classification accuracy among others for Left, Right and Straight motor imagery signal classification. The proposed greedy feature selection algorithm not only selects on an average only 12-18% of the total number of features, but also enhances the classification accuracy by 4-16%.