Utilizing big data and categorical boosting modeling methodology to interpret the load-deflection behavior of steel fiber-reinforced concrete beams

Extensive research has been conducted on steel fiber reinforced concrete (SFRC) beams, yet studies focused on predicting their load-deflection curves using machine learning (ML) remain limited. This study addresses these gaps by employing advanced ML techniques—random forest, extreme gradient boosting, and CatBoost (Categorical Boosting)—to predict SFRC beam behavior. Using a dataset of 358 beams, rigorous data preprocessing ensured integrity. Hyperparameters were optimized via grid search and cross-validation. The CatBoost model excelled, achieving low error metrics and high correlation coefficients (92.7–95.1%). Analysis identified beam span and fiber tensile strength as key predictors of critical displacements (yield, ultimate, and failure) and loads (yield, ultimate, and failure). Findings indicate beam span (L) significantly influences critical displacements and loads due to its impact on shear and moment values. Fiber tensile strength is crucial for predicting ultimate and failure displacements, especially in later loading stages. Although fiber volume fraction has a lower impact on critical displacement prediction, it ranks higher for critical load prediction, showing variable influence across prediction types. Fiber type ranks low in importance for both displacement and load predictions. This study demonstrates the potential of ML techniques, particularly CatBoost, in advancing the prediction and optimization of SFRC beam designs, offering valuable insights for modern infrastructure development.