How many clusters exist? Answer via maximum clustering similarity implemented in R

Abstract

Finding the number of clusters in a data set is considered as one of the fundamental problems in cluster analysis. This paper integrates maximum clustering similarity (MCS), for finding the optimal number of clusters, into R statistical software through the package MCSim. The similarity between the two clustering methods is calculated at the same number of clusters, using Rand [Objective criteria for the evaluation of clustering methods. J Am Stat Assoc. 1971;66:846–850.] and Jaccard [The distribution of the flora of the alpine zone. New Phytologist. 1912;11:37–50.] indices, corrected for chance agreement. The number of clusters at which the index attains its maximum with most frequency is a candidate for the optimal number of clusters. Unlike other criteria, MCS can be used with circular data. Seven clustering algorithms, existing in R, are implemented in MCSim. A graph of the number of clusters vs. clusters similarity using corrected similarity indices is produced. Values of the similarity indices and a clustering tree (dendrogram) are produced. Several examples including simulated, real, and circular data sets are presented to show how MCSim successfully works in practice.