Machine learning-based prediction of shear capacity in corroded reinforced concrete beams

Creative Commons License

Akkaya H. C., Alacalı S.

Computers and Concrete, vol.38, no.4, pp.779-818, 2026 (Peer-Reviewed Journal)

Abstract

In this study, the reliability of machine learning (ML) models for predicting the shear capacity of reinforced concrete (RC) beams exposed to corrosion was investigated. A dataset involving 204 experimental results collected from the literature was constructed for this investigation. The prediction performance of ML models was evaluated for the training and testing datasets. The mean value (MV), standard deviation (STD), root mean square error (RMSE), coefficient of determination (R²), and coefficient of variation (COV) were utilized to compare the prediction performance of ML models. It was concluded this comparison that overfitting and underfitting problems were overcome in the predictive performance of ML models. Furthermore, another conclusion from the ML model’s prediction results was that the predictions obtained using ML models were accurate and robust according to the statistical indicators mentioned above. The new equation, which was obtained with the help of the Gene Expression

Programming (GEP) model, was proposed. Moreover, Extreme Gradient Boosting (XGBoost) was selected as the best model out of eleven ML models. The MV, STD, RMSE, R² and COV values of the XGBoost model for testing dataset are 1.028, 0.116, 13.725, 0.965 and 0.113, respectively. The prediction results of the XGBoost was superior to those of the five analytical models available in the literature. Finally, SHapley Additive Explanations (SHAP) analysis was performed on the XGBoost model.