A Neutral Mutation Operator in Grammatical Evolution

Abstract

In this paper we propose a Neutral Mutation Operator (NMO) for Grammatical Evolution (GE). This novel operator is inspired by GE’s ability to create genetic diversity without causing changes in the phenotype. Neutral mutation happens naturally in the algorithm; however, forcing such changes increases success rates in symbolic regression problems profoundly with very low additional CPU and memory cost. By exploiting the genotype-phenotype mapping, this additional mutation operator allows the algorithm to explore the search space more efficiently by keeping constant genetic diversity in the population which increases the mutation potential. The NMO can be applied in combination with any other genetic operator or even different search algorithms (e.g. Differential Evolution or Particle Swarm Optimization) for GE and works especially well in small populations and larger problems.