Automated dimensionality reduction in EEG based Brain Computer Interface

A simple method is developed for selecting effective channels and feature dimensions automatically at the training stage of BCI systems. The method is applied on feature vectors constructed with all the EEG channels used for recording. Performance was evaluated with EEG data which was preprocessed by band pass filtering and feature vectors constructed by band powers. The classification method used in the evaluation is k Nearest Neighbor classifier, which is sensitive to number of dimensions in the feature vectors. It was found that new method can effectively select features and reduce channels and thereby improve accuracy and efficiency of BCI systems.