Consensus functions for cluster ensembles

Abstract

The major task of clustering is to group an heterogeneous population into unknown groups based on a similarity measure. In order to enhance the robustness and the stability of unsupervised classification solutions, clustering ensembles are used; they are considered to be a powerful tool to deal with this issue. Individual clusterers consolidate the process of decision making through the concept of weighting. The aim is to determine an effective combination method that makes use of the benefits of each clusterer while avoiding its weaknesses. In this paper, we study the problem of combining multiple partitioning without accessing the original features. A genetic algorithm is proposed using three different fitness scores. Following three scenarios: Object Distributed Clustering, Feature Distributed Clustering, and Robust Centralized Clustering, the proposed consensus functions algorithm outperforms three existing ones: Cluster-based Similarity Partitioning Algorithm, HyperGraph Partitioning Algorithm and Meta-Clustering Algorithm. Copyright © 2012 Taylor and Francis Group, LLC.