Machine learning and SHAP for predicting the shear contribution of FRP strengthening in beams
TL;DR
To address the identified research gaps, this study offers a strategy for forecasting the FRP shear contribution using seven supervised ML algorithms and Shapley Additive exPlainations (SHAP)—a novel black-box interpretation approach—was used to explain the predictions.
Abstract
Machine learning (ML) methods are frequently applied for reliable prediction of the shear strength contribution of fiber-reinforced polymer (FRP) strengthened reinforced concrete (RC) beams. While offering high accuracy, prior ML models failed to interpret the rationale behind predictions. Model interpretability is crucial to capture the attention of domain experts. Thus, addressing the identified research gaps, this study offers a strategy for forecasting the FRP shear contribution using seven supervised ML algorithms (Support vector regression, Decision tree, Random forest regression, Extra tree regressor, Gradient tree boosting regression (GTRB), Extreme gradient boost (XGBoost), and categorical gradient boosting (CatBoost)). In contrast, Shapley Additive exPlainations (SHAP)—a novel black-box interpretation approach—was used to explain the predictions. The CatBoost model demonstrated high predictive accuracy for estimating the FRP shear contribution, achieving an R 2 of 0.96 and a MAPE of 9.34% on the training dataset, and an R 2 of 0.91 and a MAPE of 15.03% on the testing dataset. SHAP analysis highlighted the significant influence of input features on the FRP shear contribution ( V f ), offering valuable insights into each feature’s impact and enhancing the interpretability of the optimal ML model.