Enhancing Compressive Strength Prediction in Recycled Aggregate Concrete Using XGBoost Optimization

Abstract

The accurate prediction of compressive strength in recycled aggregate concrete (RAC) presents a significant challenge due to its heterogeneous nature. This study focuses on enhancing the predictive performance and interpretability of the XGBoost, a powerful machine-learning algorithm, for compressive strength prediction in RAC. Using GridSearchCV, hyperparameter optimization was conducted to refine the model’s parameters. Additionally, SHAP analysis was employed to provide insights into the global and local interpretations of the model. The results demonstrate significant improvements in predictive accuracy, with the tuned XGBoost model achieving higher $\mathbf{R}^{2}$ scores compared to the default model. Furthermore, reductions in Root Mean Square Error (RMSE) and Mean Absolute Error (MAE) were observed, indicating enhanced precision in predicting compressive strength values. The SHAP analyses reveal influential features impacting RAC strength, highlighting the significance of cement, water, sand, and recycled aggregate water absorption in predicting RAC compressive strength. These findings not only contribute to advancing predictive modeling techniques in concrete engineering but also highlight the efficacy of GridSearchCV in optimizing the performance of machine learning models for heterogeneous materials like RAC.