Parameter tuning for meta-heuristics

Abstract

These days meta-heuristic algorithms are gaining lot of popularity. The performance of the meta-heuristics depends upon the suitable selection of user dependent parameters. Finding the most suitable values for the parameters (fine tuning) is a challenging problem. This paper proposes a generalized strategy to find the most suitable value of any parameter for a meta-heuristic algorithm. The approach is based on the relation between algorithm’s performance and functional landscape. The proposed approach is evaluated by applying it to a recent meta-heuristic algorithm, Gravitational Search Algorithm (GSA). The parameter α which plays a vital role in the convergence of GSA search process, is fine tuned using the proposed strategy. Obtained values of α, change the nature of gravitational coefficient G from monotonic to non-monotonic for a cluster free diversified search. The proposed strategy is tested over CEC-2015 test suite. Various statistical tests have been applied to compare the obtained results with recent variants of GSA and other state-of-the-art meta-heuristics. Results confirm that the parameters obtained using proposed approach significantly improve the results.