Comparative analysis of selection and replacement schemes for genetic programming

The selection mechanism in genetic programming has attracted attention of many researchers, as a result lexicase selection and its variants have been proposed. However, the replacement schemes are often out of the scope of most studies on genetic programming. In this paper, the effects of different selection mechanisms are studied, when they are applied to both parents’ selection and replacement. The aim of the study is to evaluate the efficiency of different parent and environment selection schemes in genetic programming. For this in the paper the four-parent selection, seven replacement are compared and two methods for selecting first parent in a pair, giving a total of 56 combinations. The experiments are performed on a set of 10 regression problems from the PMLB dataset collection, and the determination coefficient is used as a performance metric. The Mann-Whitney statistical tests and Friedman ranking is used for the analysis of the performance values. The analysis of the results show that the down-sampled lexicase selection can is an efficient mechanism, selecting current individual as one of the parents could be beneficial, and that either pair-wise comparison or selection of best of parents and offspring is an efficient replacement strategy. The results obtained in this study can be applied to improve the efficiency of solving symbolic regression problems, which often arise in the aerospace industry, for example for predicting solar panels degradation parameters, or antenna radiation patterns.