Improvement of Partitioning and Merging Phase in Chameleon Clustering Algorithm

Abstract

In view of the Chameleon algorithm is used to work with rough version hMetic library to partition the graph, each run may be slightly different as a result, difficulties in selecting the parameters of the merging phase, and similarity measure did not completely evaluate the similarity between clusters by density, the Chameleon algorithm which does not need human intervention is proposed. By introducing the recursive dichotomy, flood fill method and the quotient ϒ of cluster density on the traditional Chameleon proposed an improved scheme, also put forward a cutoff method that automatically selects the best clustering result from a modified chameleon dendrogram. Experimental results on artificial datasets and real datasets show that the Chameleon algorithm proposed in this paper has a good performance in NMI evaluation criteria, clustering accuracy and running time.