Stability of individual object in construction of voting-merged approach

Abstract

The construction of multiple partitions in consensus clustering with random initialization and various parameters values of clustering algorithm enable us to measure the stability of objects (points) in a cluster. The procedure to indicate stability is aided by the co-occurrences of pair objects (i,j) allocated to the same cluster in each partition. In this paper, we propose a voting-merged method - a combination of voting-based method and merging process. Our experiment with simulations and real datasets shows better performance for well-separated clusters and low degree of overlapping.The construction of multiple partitions in consensus clustering with random initialization and various parameters values of clustering algorithm enable us to measure the stability of objects (points) in a cluster. The procedure to indicate stability is aided by the co-occurrences of pair objects (i,j) allocated to the same cluster in each partition. In this paper, we propose a voting-merged method - a combination of voting-based method and merging process. Our experiment with simulations and real datasets shows better performance for well-separated clusters and low degree of overlapping.