A New Grammatical Evolution Based on Probabilistic Context-free Grammar

Abstract

This paper presents a new grammatical evolution (GE) that generates automatic program under favor of probabilistic context-free grammar. A population of individuals is evolved under genotypic integer strings and a mapping process is utilized to translate from genotype (i.e., integer string) to phenotype (i.e., complete program). To efficiently handle this process, unlike the standard GE that employs a simple modulo function, the probability concept is introduced to context-free grammar, thereby choosing production rules according to assigned probabilities. Moreover, any crossover and mutation are not employed for generating new individuals. Instead, along the lines of estimation of distribution algorithms that perform search using a probabilistic model of superior individuals, a new population is created/evolved from probabilistic relationship between production rules. A comparative study on the standard GE and the proposed GE is conducted; the performance achieved by the both methods is comparable. Also, the experimental results firmly demonstrate the effectiveness of the proposed approach.