Faculty of Computing, Health and Science
School of Computer and Information Science
Understanding the behaviour of different optimisation algorithms is important in order to apply the best algorithm to a particular problem. The WFG toolkit was designed to aid this task for multi-objective evolutionary algorithms (MOEAs), offering an easily modifiable framework that allows practitioners the ability to test different features by "plugging" in different forms of transformations. In doing so, the WFG toolkit provides a set of problems that exhibit a variety of different characteristics. This paper presents a comparison between two state of the art MOEAs (NSGA-II and SPEA2) that exemplifies the unique capabilities of the WFG toolkit. By altering the control parameters or even the transformations that compose the WFG problems, we are able to explore the different types of problems where SPEA2 and NSGA-II each excel. Our results show that the performance of the two algorithms differ not only on the dimensionality of the problem, but also by properties such as the shape and size of the underlying Pareto surface. As such, the tunability of the WFG toolkit is key in allowing the easy exploration of these different features.