Mohamad Hossein Fazel Zarandi; Solmaz Ghazanfar Ahari; Nader Ghaffari-Nasab
Volume 10, Issue 27 , January 2012, , Pages 22-43
Abstract
Estimating the optimal number of clusters in an unsupervisedpartitioning of data sets has been a challenging area in recent years.These indices usually use two criteria called compactness andseparation to evaluate the efficiency of the performed clustering. Inthis paper a new separation measure for ECAS ...
Read More
Estimating the optimal number of clusters in an unsupervisedpartitioning of data sets has been a challenging area in recent years.These indices usually use two criteria called compactness andseparation to evaluate the efficiency of the performed clustering. Inthis paper a new separation measure for ECAS cluster validity index,proposed by Fazel et al. [1] is identified, which uses Jaccard distancein order to consider the whole shape of clusters. Jaccard distance usesthe size of intersection and union of fuzzy sets, giving the clustervalidity index more information about the overlap and separation ofclusters. This property results in high robustness of the proposed indexdealing with various degrees of fuzziness in comparison with ECAS.To test the efficiency of the proposed index in comparison with nineother indices existing in the literature, 15 data sets (3 existing datasetsand 12 artificial data sets) have been used. Computational resultsindicate robustness and high capability of the proposed index incomparison with previous indices