How to Select the Fittest Solution for a multi variable problem

Grasshopper Wallacei
1

I was looking for advice on how to pick solutions that are well balanced for design problems with 3 or more objectives. I see from the parallel coordinate plots there are options to select “Relative Difference Between Fitness Ranks = 0” and “Average Fitness Ranks = 0” methods. I also see there is a way to cluster the Pareto front solutions and select options from that front. The part I’m not 100% understanding is the logic behind those methods of selection. For example if I wanted to find the solution from a set that balanced its fitness for all possible objectives which of those selection methods would I use and why would I use a particular one. Or perhaps if I should be looking at all of those methods (or others) what is a good hierarchy to look at those methods and what would point me to look at one method vs the other.

I understand this is a ‘it depends’ question and apologies if its in the primer and I’ve just missed it but just looking to educate myself on what these selection tools actually mean so I can deploy them in more thoughtful ways.

Thanks ~

2

Hi Frederick, we recently published a paper that presents a framework for selecting solutions generated by a MOEA. It answers all your questions above and provides more information on how one can employ a methodical approach to selection that utilizes both subjective and objective methods of selection.

Reading it will be super beneficial to what you’re trying to do.

1 Like
3

Thank so much for sharing. Is this from the recent workshop in Europe? I really wanted to go to that (for obvious reasons) but wasn’t able to make the trek.

Anyway thanks for sharing the link and helping demystify this.

4

Nope, we published this after the workshop. We will be giving more workshops in the coming months (most likely at the beginning of 2023). Keep an eye out.

1 Like