Examination of the performance of objective reduction using correlation-based weighted-sum for many objective knapsack problems

In this paper, we show the effectiveness of an EMO (Evolutionary Multi-criterion Optimization) algorithm with objective reduction using a correlation-based weighted-sum in many objective knapsack problems. Recently many EMO algorithms are proposed for various multi-objective problems. However, it is known that the convergence performance to the Pareto-frontier becomes weak in approaches using archives of non-dominated solutions since the size of archives becomes large as the number of objectives becomes large. In this paper, we show the effectiveness of using information of correlation between objectives to construct groups of objectives. Our simulation results show that while an archive-based approach, such as NSGA-II, produces a set of non-dominated solutions with better objective values in each objective, the correlation-based weighted sum approach can produce better compromise solutions that have better minimum objective values in every objective in many objective knapsack problems.