Abstract

Metaheuristic design is an incremental and difficult task. It is usually iterative and requires several evaluations of the code to obtain an algorithm with good performance. In this work, we analyse the design of metaheuristics by detecting components which are strictly necessary to obtain a good performance (in term of solutions quality). We use a collective strategy where the information generated by a tuner is used to detect the components usefulness. We evaluate this strategy with two well-known tuners EVOCA and I-RACE to analyse which one is more suitable and provides better results to make this components detection. The goal is to help the designer either to evaluate during the design process different options of the code or to simplify her/his final code without a loss in the quality of the solutions.