Performance Evaluation of Metaheuristics
Frédéric Héliodore, Amir Nakib, Boussaad Ismail, Salma Ouchraa, Laurent Schmitt · 2017
This chapter presents the different performance measures that can be used. Any optimization algorithm with stochastic characteristics is evaluated following specific indicators such as the quality of the solutions obtained, the computational effort necessary for its execution and its robustness. Once the experimental results for different indicators are obtained, statistical analysis methods can be utilized to estimate the performance of the metaheuristics. Statistical tests provide a means to perform a comparison between several metaheuristics and to determine the reliability of the results obtained. The chapter describes a few benchmarks used in the literature which aim to compare the performance of metaheuristics. Test functions can evaluate the different characteristics of optimization algorithms. The chapter outlines the most commonly used test functions with their respective descriptions, namely the sphere function, the Rosenbrock function, the Rastrigin function, the Schwefel function, the Ackley function, the Griewank function and the Michalewicz function.