MTEEGC: A novel approach for multi-trial EEG clustering

Abstract

This paper explores multi-trial EEG (Electroencephalography signal) clustering and proposes a novel centroid-based approach for it. It firstly utilizes an improved cross correlation to measure similarities of multi-trial EEGs and then proposes an optimal EEG feature extraction to seek cluster centroids based on the improved cross correlation similarities. Finally, it leads to a novel algorithm called MTEEGC for multi-trial EEG clustering. MTEEGC yields high-quality multi-trial EEG clustering with respect to the intra-cluster compactness as well as the inter-cluster scatter. Meanwhile, it also demonstrates the superiority of MTEEGC in clustering accuracy over 10 state-of-the-art time series clustering algorithms through a detailed experimentation using standard cluster validity criteria on 5 real-world multi-trial EEG datasets. Especially, compared with the worst and the best algorithms in the 10 baseline algorithms, MTEEGC respectively achieves 36.11% and 2.53% mean improvements with clustering accuracy (i.e., RI) on 5 multi-trial EEG datasets.