Predicting the compressive strength of recycled aggregate concrete using machine learning models

Abstract

This paper investigates the compressive strength of recycled aggregate concrete (RAC) using machine learning models, specifically the Symbolic Regression (SR) and XGBoost models. The dataset consists of 1,047 experimental samples collected from 40 published studies, allowing for accurate prediction of compressive strength based on parameters such as cement content, water, fine aggregates, and recycled aggregates. The results show that the XGBoost model achieved high accuracy with an MAE of 4.65 MPa and an RMSE of 7.61 MPa. The paper also analyzes the influence of these parameters on compressive strength using the SHAP method, emphasizing the importance of understanding the correlations between variables to optimize recycled concrete design.