Performance evaluation of distance metrics in the clustering algorithms

Abstract

Distance measures play an important role in cluster analysis. There is no single distance measure that best fits for all types of the clustering problems. So, it is important to find set of distance measures for different clustering techniques on datasets that yields optimal results. In this paper, an attempt has been made to evaluate ten different distance measures on eight clustering techniques. The quality of the distance measures has been computed on basis of three factors: accuracy, inter-cluster and intra-cluster distances. The performance of clustering algorithms on different distance measures has been evaluated on three artificial and six real life datasets. The experimental results reveal that the performance and quality of different distance measures vary with the nature of data as well as clustering techniques. Hence choice of distance measure must be done on basis of dataset and clustering technique.