Enhancing concrete strength for sustainability using a machine learning approach to improve mechanical performance.

Abstract

The construction industry faces growing pressure to adopt sustainable practices due to the environmental burden of concrete waste and the overuse of natural resources. One promising solution is the use of recycled concrete aggregate (RCA) as a partial or full replacement for natural aggregates. However, the inconsistent performance of RCA concrete due to differences in source material, composition, and mix design poses challenges for its widespread adoption. This study leverages machine learning (ML) to predict the mechanical performance of RCA concrete and identify the key variables influencing its strength. A robust dataset of 583 samples was compiled from the literature, featuring 10 input parameters and two key outputs: compressive strength (Fc) and split tensile strength (STS). Three ML models Extreme Gradient Boosting (XGBoost), Decision Tree, and K-Nearest Neighbors (KNN) were developed and evaluated using metrics such as R², RMSE, MAE, and MAPE. Among the models tested, XGBoost demonstrated the best performance, achieving test R² values of 0.86 for Fc and 0.88 for STS, with RMSEs of 8.32 MPa and 0.55 MPa, respectively. Decision Tree followed with moderate accuracy, while KNN showed limited predictive power. To understand feature influence, SHAP analysis was conducted, revealing the water-to-cement ratio and cement content as the most critical factors impacting strength. By integrating ML with recycled material use, this study presents a reliable predictive approach for RCA-based concrete performance offering practical insights to engineers and aiding in the transition toward greener construction solutions.