Comparison of Different Approaches for Adapting Mutation Probabilities in Genetic Algorithms

Abstract

Traditionally in Genetic Algorithms, the mutation probability parameter maintains a constant value during the search. However, an important difficulty is to determine a priori which probability value is the best suited for a given problem. In this paper we compare three different adaptive algorithms that include strategies to modify the mutation probability without external control. One adaptive strategy uses the genetic diversity present in the population to update the mutation probability. Other strategy is based on the ideas of reinforcement learning and the last one varies the probabilities of mutation depending on the fitness values of the solution. All these strategies eliminate a very expensive computational phase related to the pre-tuning of the algorithmic parameters. The empirical comparisons show that if the genetic algorithm uses the genetic diversity, as the strategy for adapting the mutation probability outperforms the other two strategies.