Weighted Co-clustering Based Clustering Ensemble

Consensus clustering has emerged as an important elaboration of classical clustering problem that improves quality and robustness in clustering by optimally combining the results of different clustering process. In this paper we propose a new method of arriving at a consensus clustering. We assign confidence score to each partition in the ensemble and compute weighted co-association for each pair of objects. In order to derive the consensus clustering from co-association matrix, we consider two cases of co-clustering based clustering technique to group the rows and columns simultaneously. The objective is to derive as many as homogeneous blocks as possible. The use of co-clustering based clustering technique captures the transitive relationship. We show empirically that for benchmark datasets, for both cases of our technique yields better consensus clustering than other major algorithms.