There is no such thing as "logical reasons" for most evolutional approaches. In general - creating some "custom", "personal" implementation of the evolutional method is not a good idea. There are dozens of developed methods, tested, evaluated, criticised and corrected. Assuming, that one idea in form of "oh, I will do this that way" will be better than those heavily discussed methods of scientists is rather naive. What is even more important - such approaches are not really a "true" machine learning methods - these are just "fuzzy" (not in the mathematical sense) heuristics to avoid complete random guessing. And as such - for almost all of them there is no true mathematics behind it. There are at least three main ways:
- If you want a "logical reasons" then select some existing, mathematically correct model for your problem, maybe some Gaussian Process, Multi-Armed Bandit based method, etc.If you have limited possibility of actually testing and evaluting (which seems to be the case as you need an user input) this looks like the best choice - so no evolution.
- If you have some time to spare, and you can perform some experiments - then select one of existing evolutional approaches which seems based suited for your needs - hundreads of scientists already faced similar issues - just google for their papers
- if you have huge amounts of time then think about your own strategy, and test it, evaluate, modify - this is very complex (and expensive) process, which does not guarantee any results (but will give you some answers regarding your intuition)
I am not a specialist in the field of evolutionary algorithms, but I've already seen very similar models and algorithms to the one outlined here, so it is rather probable that you can find papers with evaluation of such approaches and some valuable discussion sections.