Prediction and Interpretability Analysis of Concrete Tensile Strength Using Machine Learning

Abstract

Fiber-Reinforced Concrete (FRC) is used for its improved tensile performance and crack resistance. However, accurate prediction of split tensile strength remains challenging due to the nonlinear influence of mix parameters. This study presents a machine learning–based approach for predicting the split tensile strength of FRC using experimental data. Experimental data from various FRC mix designs were used to develop predictive models using advanced machine learning techniques: Extreme Gradient Boosting (XGBoost), Gaussian Process Regression (GPR), Support Vector Regression (SVR), and Random Forest (RF). Among these models,XGBoost exhibited exceptional performance, attaining R^2 =0.97, RMSE = 0.0313, and MAE = 0.0253 during the training phase, and R^2 =0.88, RMSE = 0.0836, and MAE = 0.0665 in the testing phase. To further interpret the model predictions, sensitivity analysis was conducted using Shapley Additive exPlanations (SHAP). The SHAP-based analysis revealed that fiber content and ultimate load are the most influential parameters affecting tensile strength, followed by curing age and water–cement ratio. The results highlight the effectiveness of the XGBoost model in predicting the tensile strength of Concrete and provide valuable insights into the relative importance of key input variables. This study demonstrates that machine learning-based predictive frameworks can serve as reliable tools for optimizing material design and reducing dependence on extensive experimental testing, thereby supporting the development of more efficient and sustainable construction materials.