Weighting Cluster Ensembles in Evidence Accumulation Clustering
EKDB&W 2005 EKDB&W is part of 12th Portuguese Conference on Artificial Intelligence - EPIA 2005, Covilha, Portugal, (December 2005)Abstract
We explore the idea of evidence accumulation (EAC) for combining the results of multiple clusterings. The EAC paradigm combines the information existent in n partitions into a co-association matrix (similarity matrix) based on pairwise associations, where each partition has an identical weight in the combination process. The final data partition is obtained by applying a clustering algorithm over this coassociation matrix. In this paper we propose the idea of weighting differently the partitions (WEAC). Each partition contributes differently in a weighted co-association matrix depending on the quality of the partitions, as measured by internal and relative validity indices. Based on experimental results in synthetic and real data sets, the weighting of the partitions (WEAC), generally leads to a better performance than EAC. The evaluation of results is based on a consistency index between the combined partition and the ideal data partition taken as ground truth.