A Parameter Tuning Methodology for Metaheuristics Based on Design of Experiments

Many parameters have to be tuned for any metaheuristics. Parameter tuning may permit a superior flexibility and robustness , but requires a careful initialization. Those parameters may have a large influence on the efficiency and effectiveness of the search. The optimal values for the parameters m ainly depend on the problem. In order to let a project to be replicated, a standard procedure as a methodology is required. In this paper, a parameter tuning methodology for metaheuristics based on design of experiments is proposed. The proposed methodolog y comprises five phases, namely, Problem Characteristics Screening, Clustering, Parameter Screening, Response Surface Modeling and Optimization. The proposed methodology is applied to the Ant Colony System algorithm for solving 47 traveling salesman problem instances. For validation of the proposed methodology, the different alternative approaches for parameter tuning are compared and it is concluded that, the methodology presents better results than the other alternative approaches.