Determining the number of clusters in CROKI2 algorithm
Malika Charrad, Mohamed Ben Ahmed, Yves Lechevallier, Gilbert Saporta · INRIA a CCSD electronic archive server · 2009
One of the major problems in clustering is the needof specifying the optimal number of clusters in someclustering algorithms. Some block clusteringalgorithms suffer from the same limitation that thenumber of clusters needs to be specified by a humanuser. This problem has been subject of wide research.Numerous indices were proposed in order to findreasonable number of clusters. In this paper, we aim toextend the use of these indices to block clusteringalgorithms. Therefore, an exami nation of some indicesfor determining the number of clusters in CROKI2algorithm is conducted on both real data extractedfrom Metz web site and synthetic data sets beinggenerated according to a methodology that will beexplained later. The purpose of the paper is to test theperformance and ability of some indices to detect theproper number of clusters on rows and columns and tocompare our new index with some other indexes.