Clustering based leaders' selection in multi-objective evolutionary algorithms

Al Moubayed, N; Petrovski, A; McCall, J

Abstract

Clustering-based Leaders Selection (CLS) is a novel leaders selection technique in multi-objective evolutionary algorithms. Clustering is applied on both the objective and solution spaces whereby each individual is assigned to two clusters; one in the objective space and the other in the solution space. Mapping between clusters in both spaces is then applied to recognize regions with potentially better solutions. A leaders archive is used where a representative of each cluster in the objective and solution spaces is stored. The results of applying CLS integrated with NSGAII on seven standard multi-objective problems, show that clustering based leaders selection NSGAII (NSGAII/C) is highly competitive comparing with the original algorithm.

Citation

Al Moubayed, N., Petrovski, A., & McCall, J. (2011, July). Clustering based leaders' selection in multi-objective evolutionary algorithms. Presented at Proceedings of the 13th annual conference companion on Genetic and evolutionary computation - GECCO '11, Dublin, Irland

• Presentation Conference Type: Conference Paper (published)
• Conference Name: Proceedings of the 13th annual conference companion on Genetic and evolutionary computation - GECCO '11
• Start Date: Jul 12, 2011
• End Date: Jul 16, 2011
• Online Publication Date: Jul 12, 2011
• Publication Date: 2011-07
• Deposit Date: Jan 26, 2016
• Publisher: Association for Computing Machinery (ACM)
• Pages: 95-96
• ISBN: 9781450306904
• DOI: https://doi.org/10.1145/2001858.2001913
• Keywords: Leaders Selection, Multi-Objective Optimization, Clustering, Evolutionary Algorithm, Density Based Spatial Clustering, Principal Component Analysis.
• Public URL: https://durham-repository.worktribe.com/output/1151590