Voting-based efficient cluster ensemble fusion

Abstract

Voting-based consensus clustering is a subset of consensus techniques that makes clear the cluster label mismatch issue. Finding the best relabeling for a given partition in relation to a reference partition is known as the voting problem. As a weighted bipartite matching problem, it is frequently formulated. We propose a more generic formulation of the voting problem as a regression problem with various and multiple-input variables in this work. We demonstrate how a recently developed cumulative voting system is an exception that corresponds to a linear regression technique. We employ a randomised ensemble creation method in which an excess of clusters are randomly chosen for each ensemble partition. In order to extract the consensus clustering from the combined ensemble representation and to calculate the number of clusters, we use an information-theoretic approach. Together with bipartite matching and cumulative voting, we use it. We provide empirical data demonstrating significant enhancements in clustering stability, estimation of the real number of clusters, and accuracy of clustering based on cumulative voting. The gains are made in comparison to recent consensus algorithms as well as bipartite matching-based consensus algorithms, which struggle with the selected ensemble generation technique.