Supervised Learning Used in Automatic EEG Graphoelements Classification

The comparison of supervised (k-nearest neighbors) and unsupervised (k-means) methods for automatic classification of EEG grapholements is presented here. The resulting classes should distinguish EEG impulse artifacts, epileptic EEG, EMG activity, normal EEG and many more. The classified EEG graphoelements are visualized in the original multi-channel EEG recording by coloring the EEG graphoelements itselves according to the class they belong to. The temporal profiles of the EEG recording are plotted. The whole procedure of classification begins with adaptive segmentation of EEG graphoelements and feature extraction followed by classification. This data processing approach ends in colored graphoelements according to class directly in the EEG recording, which is suggested to the electroencephalographer for more effective multi-channel EEG analysis.