Estimation of the Number of Clusters by using Information Criterion in SOM-based Clustering

Abstract

In a process of cluster extraction by using Self-Organizing Map(SOM), information criterion such as AIC make it possible to improve the accuracy of cluster estimation. Furthermore, the number of clusters also can be estimated by using the information criterion. However, accurate estimation of the number of clusters is difficult when clusters have significantly different cluster-sizes and densities each other. In this paper, we examine the application of the information criterion to the cluster estimation process by which we try to resolve this problem.