EVOLUTIONARY ALGORITHMS PERFORMANCE EVALUATION USING RANK-BASED MULTIPLE COMPARISON PROCEDURE

摘要

Comparing the performance of different evolutionary optimization algorithms is not trivial and is an important task to understand and develop new and more effective approaches. Also, as EAs are stochastic process with unknown distribution, performance comparison must be done with the proper statistical tools and procedures. This paper proposes a generic comparison procedure to evaluate simultaneously the performance of several different EAs. The proposed comparison procedure has the advantage to be statistically consistent and adequate to evaluate performance analysis of the EAs. With the ranking scheme, it also allows a unification of the three major criteria under a unique ranking number without the use of using arbitrary artefacts such as weighting coefficients or function objective transformations.