Clustering Ensemble Based on a New Consensus Function with Hamming Distance

Abstract

Unlike classification problems, there are no well known approaches to combining multiple clusterings which is more difficult than designing classifier ensembles since cluster labels are unknown. A new algorithm is to use Hamming distance as the similarity metric to find the best partition is proposed. Also a scheme for a selective initial cluster centers by Hamming distance is used in the consensus function, which help us to find the most likely different classes of the data. Experiment results show that the algorithm is more stable, higher performance and more efficiently than other compared methods.