Evaluating the Generalization Capability of XGBoost Model in Predicting Shear Strength of Reinforced Concrete Deep Beams

الملخص

Extreme Gradient Boosting (XGBoost) is a powerful machine learning algorithm widely adopted for solving complex regression problems in structural engineering. This study investigates the capability of the XGBoost model in predicting the shear strength of reinforced concrete (RC) deep beams—a task characterized by highly nonlinear behavior and interaction between several geometric, material, and reinforcement parameters. The model was trained on a comprehensive database of 1177 experimental specimens using 25 input variables and achieved a high prediction accuracy with a coefficient of determination (R²) of 0.96, indicating excellent fitting and learning capacity. To evaluate its generalization ability, the trained model was further validated using an unseen dataset consisting of 151 RC deep beam specimens described by 11 influential variables. The prediction accuracy on this external dataset reached an R² value of 0.83, demonstrating satisfactory robustness and reliable extrapolation performance. These results highlight the efficiency and precision of the XGBoost model in capturing the complex shear behavior of deep beams and confirm its strong potential as a practical predictive tool to assist in the analysis and preliminary design of RC deep beams.