MOCS: multi-objective clustering selection evolutionary algorithm
Thomas E. Koch, Andreas Zell · 2002
In this paper, we describe a multi-objective evolutionary algorithm, that uses clustering selection and does not need any additional parameter like others. It clusters the population into a flexible number of clusters employing x-means from [Pelleg and Moore, 2000]. First, the selective fitness is assigned to clusters and in second place to individuals of clusters. We show three hybrid variants incorporating additional mechanisms from other elitist multi-objective evolutionary algorithms in order to increase selection pressure. Using the test functions from Deb's T suite (T1-T6), from Schaffer, Kursawe and Quagliarella we evaluate the performance and the quality of our approach against the most recent and performant elitist multi-objective evolutionary algorithms, NSGA2, SPEA2 and PESA2. The comparison yields promising results for region-based selection using clustering in combination with additional crowding strategies.