A Method of Two-Stage Clustering Based on Cluster Validity Measures
Ryo Ozaki, Yukihiro Hamasuna, Yasunori Endo · 2016
Two-stage clustering is constructed from generating stage and merging one. To handle a large scale of objects, an algorithm of the two-stage clustering generates a large number of clusters in the first stage and merge clusters in the second stage. A novel two-stage clustering method is proposed by introducing cluster validity measures which are used to evaluate cluster partition and determine the suitable number of clusters. The significant cluster validity measure is used in the second stage and play a role as criterion to merge clusters. The performance of the proposed method are compared with six artificial datasets and three benchmark datasets. These experiments show that several cluster validity measures, that is, trace of fuzzy covariance matrix and membership degrees based indices are effective in the proposed method and obtain better results than other indices.