THE EVALUATION OF CHF COEFFICIENT IN DETERMINING THE NUMBER OF CLUSTERS USING EUCLIDEAN DISTANCE MEASURE

There are many methods of clustering in current literature and there is possible to use the various measures of distances (resp. similarities), and therefore the resulting distributions of objects into the clusters may be different. There is no strict rule in the literature to determine which method is necessary to use and in which conditions. Also, one part of the cluster analysis is very often to determine the number of clusters. The aim of this paper is to evaluate the CHF coefficient, which is often used to determine the number of clusters. There will be used 20 artificially created files for clustering. Conditions are, that the clusters must be touched or partially overlapped. These generated files are created under the same conditions in order to consider the objective results. Based on analyses it was found, that the CHF coefficient is very successful in determining the number of clusters. In the case that the clusters are touched to each other, its success is 100 % at the generated files. In the case that the clusters are partially overlapped, its success decreases. The highest success in the case of partially overlapped clusters was 70 %. It can be concluded, that the lower rate of separation is, (i.e. the more individual clusters are overlapped), the lower is the success of this coefficient in order to determine the number of clusters.