Clustering of categoric data in medicine - Application of evolutionary algorithms

Abstract

Clustering of non-metric data sets often occurs in investigations in medicine and social science. The problem is to find suitable measures which describe similarities and, hence, are applicable to the clustering algorithm. In the present contribution we use evolutionary algorithms EA for clustering. Thereby, the similarity measures determine the respective fitness function for the EA. We consider several fitness functions and derive a new one which allows, additionally, the determination of a useful cluster number. - For the EA we use a new selection strategy combining the advantages of both the (μ, λ)- and (μ + λ)-strategy and a multiple subpopulation approach with a migration scheme following the collective learning dynamic in self-organizing maps. © Springer-Verlag 2001.