Adaptive mutation/crossover rates for genetic algorithms

Question

I've been looking into ways I can implement an adaptive mutation rate for a genetic algorithm I'm working on. I've seen an algorithm that uses the current individual's fitness and the average population fitness to calculate a mutation rate, however, I'm not sure it's very effective.

In the algorithm I've seen you do the following:

mutationRate = (bestFitness - individualFitness) / (bestFitness - averageFitness) * 0.5

Would this be a good approach or are there better?

Answer 1

The crossover and mutation rates are adapted in response to the evaluation results of the respective offspring in the next generation. Experimental results determine that the proposed scheme significantly improves the performance of genetic algorithms and outperforms previous work. With the adaptive mutation approach, you're using, an individual with high fitness corresponds to a smaller mutation probability and individuals with low fitness corresponds to a high mutation probability. This technique can efficiently protect excellent individuals, but it is easy to fall into local convergence.