Determining the Optimal Number of Clusters in Cluster Analysis

Cluster analysis is the multivariate method which objective is to classify the objects. In current literature there are many methods and many distances measures, which can be mutually combined. There is no manual and rule which would clearly identify the appropriate combination method and distance measures during clustering. Simultaneously, in cluster analysis it is often necessary to determine the optimal number of clusters in to which the objects are to be classified. The aim of this paper is to illustrate the possibilities of the process of determining the number of clusters and to evaluate selected coefficients for determining the number of clusters in combination with clustering different methods and with different distance measures. For example CHF coefficient is more suitable to be used with combination with Mahalanobis distance, where the success is higher in comparison with Euclidean distance. For example using average linkage method the success is higher by 21.88%. On the other hand, coefficient D-B is more successful while using Euclidean distance measures. In the case of Ward’s method the success is higher by 15.63%.