Abstract

Brain Computer Interfacing (BCI) is a powerful communication tool that exists between a man and a machine. Electroencephalography (EEG) based BCI is popular in the area of rehabilitation of motor disabled people. Motor Imagery (MI) EEG is the recording of brain potentials that occurs while imagining a body movement. Identification of relevant channels in a multi-channel MI EEG is necessary for a reliable MI BCI system. Reducing the large number of channels will greatly aid in reducing system complexity and improving performance. A channel selection method based on Inverse Coefficient of Variation is proposed in this work. Common Spatial Pattern is used for feature extraction utilizing the principle of sub-band CSP. Various classifier models are used for classification to identify the best performing classifier for the particular MI task. The reduced 10 channel subset yields better performance than the original 64 channel set. Enhancement in performance accuracy when the number of channels gets reduced is evident for all classifiers. The accuracy is 52.38% when the whole set of electrodes is employed, whereas accuracy increases to a maximum of 92.85% when only chosen 10 channels are used.