ABSTRACT
We propose an EEG analysis and recognition method for the right/left hand movement discrimination in a single trial. The 30 channels-EEG was recorded during the self-paced voluntary hand movement. First, we made a feature set for every trial that represents the characteristics to reflect the process of the right/left movement. It was composed of the 256 feature units including ERD, ERS amount and timing patterns of the alpha and beta rhythm and SPC, TPC over global areas. The more feature units included in feature space, the more variability between trials and subjects and the more complexity of the system. We developed the system reduction algorithm using hierarchical clustering to decrease the variability and to make more robust feature vector space. It includes grouping of high correlated features by hierarchical clustering and generating of new feature vector space with the principal components representative of the each feature group. It reduced the dimension of the feature set and replaced the feature vector space with more robust components to make the redundancy or noise of system minimized. We estimated the performance of the reduced system to find efficient.
