A Comparative Study of Genetic Programming Variants

Genetic programming tends to optimize complicated structures producing human-competitive results; therefore, it is applied to a wide range of problems such as classification and regression. This work experimentally performs a comparative study of Genetic programming variants, namely gene expression, grammatical evolution, Cartesian, multi-expression programming, and stacked-based as general regression and classification solvers. The analyses will help to understand the strengths of each variant and identify the relative performance of variants that stand relative to each other for the given problem domains. To determine the performance difference between selected GP variants, hyperparameter tuning was performed on each GP variant for each dataset to minimize the performance difference due to implementation. A total of 11 datasets were used in the experiments, seven from the regression benchmark suite, and four from the classification. The obtained results indicate that the choice of Genetic programming variant has an impact on the performance of regression and classification problems. Multi-expression programming exhibits outstanding performance as a regression and classification solver which scales graciously with problem size and complexity whereas other variants were problem-dependent. Future work could consider implementing a multi-expression paradigm with other Genetic programming variants such as grammatical evolution and gene expression programming.