Prediction of splitting tensile strength of fiber-reinforced recycled aggregate concrete utilizing machine learning models with SHAP analysis

The infrastructure industry utilizes a significant number of natural resources and produces a lot of construction waste, both of which have negative environmental effects. As a solution, recycled aggregate concrete has emerged as a practical substitute. Predicting strength accurately is essential for cutting design time and expenses while limiting material waste from numerous mixing tests. Machine learning methods tackle structural engineering issues, including the prediction of Splitting Tensile Strength (STS). In this study, used four novel machine learning models such as Random Forest Regression (RFR), Extreme Gradient Boosting (XGBoost), Gradient Boosted Regression Trees (GBRT), and Bagging Regressor (BR) with grid search for hyperparameter tuning to forecast the splitting tensile strength of fiber-reinforced recycled aggregate concrete (FRRAC). The machine learning models demonstrated high reliability in predicting splitting tensile strength, including robust values for R-squared (R 2 ), Root Mean Square Error (RMSE), Mean Absolute Percentage Error (MAPE) and Mean Absolute Error (MAE). The prediction performance of the GBRT models showed the greatest R 2 value of 0.95 during the training stage and R 2 value of 0.83 during the testing phase. The XGBoost, RFR and BR models found R-square values were 0.822, 0.781 and 0.824 at the testing phase, respectively. Moreover, the RFR, BR, GBRT, and XGBoost model RMSE values were found to be 0.333, 0.298, 0.276, and 0.3004 at the testing phase, respectively, where the GBRT model RMSE value was found to be good. The GBRT model showed the lowest uncertainty value of both phases, with values of 0.619 and 0.597 for the training and testing phases, respectively. Furthermore, SHapley Additive exPlanations (SHAP) analysis found that CR, and additional of Fiber were the most influential input features and replacement percentage of CR (%) and RCA Absorption capacity (%) inputs had the lowest impact of Fiber-Reinforced Recycled Aggregate Concrete for predicting splitting tensile strength. These results indicate that the suggested technique can significantly contribute to sustainable construction practices by precisely predicting splitting tensile strength. • Novel application of machine learning models to predict the split tensile strength of FRRAC. • GBRT model achieved the highest accuracy compared to the other models. • SHAP revealed that cement ratio and fiber addition were the most significant features impact STS prediction.