Internal versus External cluster validation indexes

One of fundamental challenges of clustering is how to evaluate results, without auxiliary information. A common approach for evaluation of clustering results is to use validity indexes. Clustering validity approaches can use three criteria: External criteria (evaluate the result with respect to a pre-specified structure), internal criteria (evaluate the result with respect a information intrinsic to the data alone). Consequently, different types of indexes are used to solve different types of problems and indexes selection depends on the kind of available information. That is why in this paper we show a comparison between external and internal indexes. Results obtained in this study indicate that internal indexes are more accurate in group determining in a given clustering structure. Six internal indexes were used in this study: BIC, CH, DB, SIL, NIVA and DUNN and four external indexes (F-measure, NMIMeasure, Entropy, Purity). The clusters that were used were obtained through clustering algorithms K-means and Bissecting-K- means.