Combining multiple partitions created with a graph-based construction for data clustering

摘要

This paper focusses on a new clustering method called evidence accumulation clustering with dual rooted prim tree cuts (EAC-DC), based on the principle of cluster ensembles also known as “combining multiple clustering methods”. A simple weak clustering algorithm is introduced based upon the properties of dual rooted minimal spanning trees and it is extended to multiple rooted trees. Co-association measures are proposed that account for the cluster sets obtained by these methods. These are exploited in order to obtain new ensemble consensus clustering algorithms. The EAC-DC methodology applied to both real and synthetic data sets demonstrates the superiority of the proposed methods.