%0 Conference Paper %1 AUSDM2005 %A Duarte, F.Jorge F. %A L.N.Fred, Ana %A Lourenco, Andre %A Rodrigues, M. Fatima C. %B 4th Australasian Conference on Knowledge Discovery and Data Mining (AusDM05) %C Sydney %D 2005 %K Clustering_Ensembles Combining_Multiple_Partitions Validity_Indices Weighting %T Weighted Evidence Accumulation Clustering %X We explore evidence accumulation (EAC) for combining clustering ensembles. According to EAC, a voting mechanism, where each partition has an identical weight in the combination process, is used to combine N partitions into a co-association matrix. This matrix is constructed based on co-occurrences of pairs of patterns in the same cluster. A final data partition is ob-tained by applying a clustering algorithm over this co-association matrix. In this paper we propose the idea of weighting the partitions differently (WEAC). Depending on the quality of the partitions, measured by internal and relative validity indices, each partition contributes differently in a weighted co-association matrix. We propose two ways of weighting each partition: SWEAC, using a single validation index, and JWEAC, using a committee of indices. The new approach is evaluated experimentally on synthetic and real data sets, in comparison with the EAC technique and the graph-based combination methods by Strehl and Gosh, leading in general to better results.

@inproceedings{AUSDM2005, abstract = {We explore evidence accumulation (EAC) for combining clustering ensembles. According to EAC, a voting mechanism, where each partition has an identical weight in the combination process, is used to combine N partitions into a co-association matrix. This matrix is constructed based on co-occurrences of pairs of patterns in the same cluster. A final data partition is ob-tained by applying a clustering algorithm over this co-association matrix. In this paper we propose the idea of weighting the partitions differently (WEAC). Depending on the quality of the partitions, measured by internal and relative validity indices, each partition contributes differently in a weighted co-association matrix. We propose two ways of weighting each partition: SWEAC, using a single validation index, and JWEAC, using a committee of indices. The new approach is evaluated experimentally on synthetic and real data sets, in comparison with the EAC technique and the graph-based combination methods by Strehl and Gosh, leading in general to better results.}, added-at = {2009-10-25T21:22:32.000+0100}, address = {Sydney}, author = {Duarte, F.Jorge F. and L.N.Fred, Ana and Lourenco, Andre and Rodrigues, M. Fatima C.}, biburl = {https://www.bibsonomy.org/bibtex/2474d5092b418dc050e8aaa91b3510440/alourenco}, booktitle = {4th Australasian Conference on Knowledge Discovery and Data Mining (AusDM05)}, interhash = {040a4ae4662f1ea0561eaef5804835f6}, intrahash = {474d5092b418dc050e8aaa91b3510440}, keywords = {Clustering_Ensembles Combining_Multiple_Partitions Validity_Indices Weighting}, month = {December}, timestamp = {2009-10-25T21:22:32.000+0100}, title = {Weighted Evidence Accumulation Clustering}, year = 2005 }