A Study on Automatic Estimation of The Number of Clusters based on Hierarchical Clustering

Abstract

Clustering is a method of dividing data into groups called clusters based on the degree of similarity or dissimilarity between the data. When performing clustering, the number of clusters may have to be specified in advance. However, estimating the appropriate number of clusters is difficult. In non-hierarchical clustering, many methods for automatically estimating the number of clusters, such as an extension of the k-means method and the X-means method, have been studied. As a method for automatically estimating the number of clusters in hierarchical clustering, a cluster validity index and a method using a Gaussian mixture model based method have been studied. In this study, we conducted a comparative experiment between automatically estimating the number of clusters by hierarchical clustering using the cluster validity index proposed in a previous study and the method using Xie-Beni’s index. The method proposed in the previous study obtains better results than Xie-Beni’s index based method.