Electrode selection based on k-means for motor activity classification in EEG


R. Lemuz-López, W. Gómez-López, I. Ayaquica-Martínez, C. Guillén-Galván



We present an algorithm for electrodes selection associated with motor imagery activity. The algorithm uses a clustering technique called k-means to form groups of sensors and selects the group corresponding to the highest correlation activity. Then, we evaluate the selected electrodes computing the classification index using the projective decomposition called common spatial patterns and a linear discriminant method in a left hand vs right foot motor imagery classification task. This approach significantly reduces the number of electrodes from 118 to 35 while improving the classification accuracy index.