scientific article; zbMATH DE number 4082764

From MaRDI portal
Publication:3811541





[For the entire collection see Zbl 0648.00013.] We show by counterexample that the solution set of the basic convergence theorem for the fuzzy ISODATA clustering algorithms is not the set of local minima. Furthermore, we give a theorem which shows that the result for the example holds more generally. In the process we also show that the algorithms are potentially subject to ``cluster-splitting and suffer in that inspection of a given solution does not necessarily show the presence or absence of compact well-separated clusters. Finally, the nature of the counterexamples indicates that there are problems for which the domains of attraction to an undesirable solution can be large.











This page was built for publication:

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q3811541)