Permutation Tests for Metaheuristic Algorithms

Many metaheuristic approaches are inherently stochastic. In order to compare such methods, statistical tests needed. However, choosing an appropriate test is not trivial, given that each has some assumptions about the distribution of underlying data must be true before it can used. Permutation (P-Tests) with minimal number assumptions. These simple, intuitive and nonparametric. this paper, we argue researchers in field metaheuristics adopt P-Tests their algorithms. We define two statistic then present algorithm uses them compute p-value. The proposed process used 5 algorithms on 10 benchmark functions. resulting p-values compared widely tests. results show P-test generally consistent classical tests, but more conservative few cases.