Stacked ensemble and SHAP-based approach for predicting plastic rotational capacity in RC columns

The accurate estimation of plastic rotational capacity in reinforced concrete (RC) elements is essential for performance-based seismic design and structural safety assessments. In this study, an extensive experimental database, comprising 258 rectangular and 151 circular RC column specimens, was compiled based on open data available and used to train machine learning models for predicting this parameter. Three algorithms, i.e. Support Vector Regression (SVR), Random Forest (RF), and Extreme Gradient Boosting (XGBoost), were implemented and optimized using grid search within a nested cross-validation framework. The predictive performance was evaluated by averaging the coefficient of determination (R 2 ) across five outer folds, while final accuracy was assessed on the test set using both R 2 , the Mean Absolute Error (MAE), the root mean squared error (RMSE), and the mean absolute percentage error (MAPE). Model interpretability was improved using SHAP (SHapley Additive exPlanations) analysis, which quantified the influence of input parameters on predictions. Finally, a stacking ensemble model was developed to integrate the strengths of the individual regressors. The proposed methodology demonstrates increased accuracy and robustness in predicting the plastic rotational capacity of both circular and rectangular RC columns, providing a valuable tool for seismic assessment and structural reliability analysis.