The Impact of Tournament Size on the Performance of Evolutionary Multi-Objective and Many-Objective Algorithms

Tournament selection is an operator used in many evolutionary multi-objective (EMO) algorithms. The tournament size is a parameter of tournament selection, and it is generally considered that increasing the tournament size improves conver-gence while reducing diversity. To the best of our knowledge, there is no literature that analyzes the relationship between EMO algorithms and tournament size, and it is not clear what impact increasing the tournament size has on EMO algorithms. In many past studies, the tournament size has been fixed at 2 without a clear rationale. However, the likelihood that 2 is the optimal tournament size is low. Therefore, this paper investigates the performance of EMO algorithms using tournament selection when the tournament size is changed for various benchmark problems. Through experiments, we demonstrate that increasing the tournament size beyond 2 yields better results for many benchmark problems. Additionally, we experimentally show that AGEMOEAIYITS, which combines AGEMOEAII and Increase Tournament Selection, where the tournament size dynamically increases with each generation, performs well on many bench-mark problems.