Predicting the Compressive Strength of Recycled Concrete Using Ensemble Learning Model

This research proposes a stacking machine learning method to accurately predict the compressive strength of recycled concrete. The model integrates eXtreme Gradient Boosting (XGBoost), Extra Trees (ET), Decision Tree (DT), and Linear Regression (LR) models, aiming to maximize the prediction accuracy of concrete compressive strength. The model was evaluated using a combination of 63 self-made recycled concrete datasets and 1030 concrete datasets from the UCI Machine Learning Repository. Through optimization based on SHAP values, the new model achieved statistical metrics of RMSE = 1.969, MAE = 1.113, and R2 = 0.987. The comparison and analysis with the existing work show that this method has excellent performance. Additionally, the feature importance analysis based on SHAP values identified the key input variables affecting concrete compressive strength and improved the model’s prediction performance.