A new reduced-length genetic representation for evolutionary multiobjective clustering

Abstract

The last decade has seen a growing body of research illustrating the advantages of the evolutionary multiobjective approach to data clustering. The scalability of such an approach, however, is a topic which merits more attention given the unprecedented volumes of data generated nowadays. This paper proposes a reduced-length representation for evolutionary multiobjective clustering. The new encoding explicitly prunes the solution space and allows the search method to focus on its most promising regions. Moreover, it allows us to precompute information in order to alleviate the computational overhead caused by the processing of candidate individuals during optimisation. We investigate the suitability of this proposal in the context of a representative algorithm from the literature: MOCK. Our results indicate that the new reduced-length representation significantly improves the effectiveness and computational efficiency of MOCK specifically, and can be seen as a further step towards a better scalability of evolutionary multiobjective clustering in general.