New approach to determine the optimal number of clusters K in unsupervised classification

Abstract

One of the most common problems in the field of information processing is the definition of the optimal number of clusters when using unsupervised classification. The wrong choice of the number of clusters leads to insignificant and inconsistent results. In this paper, we present a simple automatic technique for estimating the optimal number of clusters when using a classification algorithm based on k-means. The proposed algorithm is based on detecting the immutable clusters during the different experiments and takes them as a reference in order to specify the best classification.