LAUSR

Abstract Background Brain Computer Interface (BCI) is explored as a new technology for communicating with computer over past few decades. It uses signals collected from brain to communicate, control or instruct computer or electronic devices. Analyzing the signals collected is most important task. If the collected data contains overlapping classes, directly applying classification techniques is inefficient. Methodology To examine and analyze the data, clustering can be useful to exploit information about dispersion of different classes. In this paper, we propose an agglomerative method for clustering high dimensional EEG signal data using multi prototype approach. This bi-phase algorithm chooses appropriate representatives in first phase, and combines them in second phase. We use squared error clustering in first phase to produce multiple prototypes located in highly dense region. A new combination measure is also proposed using fuzzy logic, to evaluate degree of prototypes can be combined. Results The proposed algorithm has same run time complexity as k-means. The proposed algorithm cluster the data with complex shapes and disperse. Experiments, carried out using synthetic and real datasets, demonstrate the performance of the proposed method in terms of accuracy and time.