Symbolic regression based on genetic programming to predict the permeability of pervious concrete

Abstract

In this study, the authors present an approach to construct a simple and easy-to-apply prediction function to predict the permeability of pervious concrete. The data set used in the study includes 267 experimental samples, each sample includes input factors such as sand amount, aggregate size, water-cement ratio, and aggregate-cement ratio, and the output factor is the permeability of concrete. We applied the Operon model, one of the most effective symbolic regression models based on genetic programming, to construct a function to predict the permeability of previous concrete. This function achieves high accuracy when compared to one of the best black-box models, PSO-XGB. The accuracy of both of these methods exceeds 0.9, but the symbolic regression function clearly shows the advantage that the function is expressed explicitly and the application also becomes simpler.