Termination criterion for Genetic Algorithm when used in context of feature selection??

I have tried for 50 iterations but on running the Matlab code the best fitness value of all iterations is coming out to be different at different times. How will I decide which will be the best features in such a condition.

Getting different best feature set for same number of iterations. How results should be interpreted so that I can come to a Termination criterion??

Answer 1

[Prateekshya]

Hello Purti,

I understand that you are getting different output in different runs of Genetic Algorithm. This is a common behavior due to the stoachastic nature of GA which gives you near-optimal (not exactly optimal) solutions. Here are some strategies to make the results more consistent:

1. Run GA multiple times and take the average of all the best fitness values.
2. Increase population size and number of generations which will explore a larger solution space and give a better result.
3. Change and experiment the type of crossovers and mutations you are using. Please follow this link to know more about it: https://www.mathworks.com/help/gads/genetic-algorithm-options.html
4. Use elitism to ensure that the best solutions are carried over to the next generation.

I hope this helps!

Thank you.