Clustering Criteria in Multiobjective Data Clustering

Julia Handl¹ and Joshua Knowles²

¹Manchester Business School, University of Manchester, UK
julia.handl@mbs.ac.uk

²School of Computer Science, University of Manchester, UK
j.knowles@manchester.ac.uk

Abstract. We consider the choice of clustering criteria for use in multiobjective data clustering. We evaluate four different pairs of criteria, three employed in recent evolutionary algorithms for multiobjective clustering, and one from Delattre and Hansen’s seminal exact bicriterion method. The criteria pairs are tested here within a single multiobjective evolutionary algorithm and representation scheme to isolate their effects from other considerations. Results on a range of data sets reveal significant performance differences, which can be understood in relation to certain types of challenging cluster structure, and the mathematical form of the criteria. A performance advantage is generally found for those methods that make limited use of cluster centroids and assess partitionings based on aggregate measures of the location of all data points.

LNCS 7492, p. 32 ff.

Full article in PDF | BibTeX

lncs@springer.com
© Springer-Verlag Berlin Heidelberg 2012