A dynamic approach for parameter tuning in genetic algorithm using crossover and mutation ratios

Abstract

Genetic algorithm uses the natural selection process for any search process. It is an optimization process where integration among different vital parameters like crossover and mutation plays a major role. The parameters have an impact on the algorithm by their probabilities. In this paper we would review the different strategies used for the selection of crossover and mutation ratios and suggest a dynamic approach for modifying the ratios during runtime. We start with a mutation ratio 0% and crossover ratio 100% where the mutation ratio slowly increases and the crossover ratio decreases (MICD). The final mutation ratio will be 0% and crossover ratio will be 100% at the end of the search process. We also do the reverse process of considering the mutation ratio to be maximum and crossover ratio to be minimum and slowly decrease the mutation ratio and increase the crossover ratio (MDCI). We compare the proposed method with two pre-existing parameter tuning methods and found that this dynamic approach of incrementing the mutation and decrementing the crossover value was more effective when the size of the population was large.