Abstract:
Gene Expression Programming (GEP) is a new member of evolutionary algorithm family, and it is an artificial genotype/phenotype system. Aiming to discover compact mathematical functions for function finding, this study analyzes the factors that greatly affect the efficiency of GEP, proposes the fitness function with pressure parameter, and implements a naïve gene expression programming (NGEP) for compact function mining tasks. Extensive experiments show that the proposed fitness function with compact pressure can automatically mine the compact functions as well as an alternative strategy to find compact results, and NGEP boosts the convergence speed by 21.7% than GEP, in addition, the results are more understandable than that are found by GEP. Compared with the evolution system without compact pressure, the average compact rate are 31.2% in GEP and 42.5% in NGEP, respectively, which shows that NGEP is easier to find compact results than GEP and the results are more comprehensive than traditional GEP.