Clustering Nominal and Numerical Data: A New Distance Concept for a Hybrid Genetic Algorithm

As intrinsic structures, like the number of clusters, is, for real data, a major issue of the clustering problem, we propose, in this paper, CHyGA (Clustering Hybrid Genetic Algorithm) an hybrid genetic algorithm for clustering. CHyGA treats the clustering problem as an optimization problem and searches for an optimal number of clusters characterized by an optimal distribution of instances into the clusters. CHyGA introduces a new representation of solutions and uses dedicated operators, such as one iteration of K-means as a mutation operator. In order to deal with nominal data, we propose a new definition of the cluster center concept and demonstrate its properties. Experimental results on classical benchmarks are given.