Consensus Clustering with Robust Evidence Accumulation

Consensus clustering methodologies combine a set of partitions on the clustering ensemble providing a consensus partition. One of the drawbacks of the standard combination algorithms is that all the partitions of the ensemble have the same weight on the aggregation process. By making a differentiation among the partitions the quality of the consensus could be improved. In this paper we propose a novel formulation that tries to find a median-partition for the clustering ensemble process based on the evidence accumulation framework, but including a weighting mechanism that allows to differentiate the importance of the partitions of the ensemble in order to become more robust to noisy ensembles. Experiments on both synthetic and real benchmark data show the effectiveness of the proposed approach.