The Development and Optimization of Machine Learning Models for Predicting the Shear Capacity of Corroded Reinforced Concrete Beams

Abstract

The deterioration of steel reinforcement through corrosion triggers cracking and loss of concrete cover, ultimately weakening the structure’s strength and ductility. In practical design and assessment, it is vital to precisely quantify the shear capacity of corroded reinforced concrete beams (CRCBs). In this paper, machine learning (ML) models are developed to predict the shear capacity of CRCBs, including kernel ridge regression (KRR), K-nearest neighbors (KNN), decision trees (DT), random forest (RF), gradient-boosted regression trees (GBRT), and extreme gradient boosting (XGBoost). A total of 408 data entries on the shear strength of CRCBs under different corrosion conditions were collected to establish an extensive database. The reliability of the proposed ML models is examined by contrasting their outputs with the experimental data. The XGBoost model demonstrated superior predictive capability, achieving an R2 value of 0.994 and outperforming all other tested models, including RF, GBRT, and DT. The Shapley Additive Explanations (SHAP) algorithm is adopted to reveal the contribution of each input feature to the predicted shear capacity of CRCBs. The interpretive SHAP results show that the ultimate shear capacity of CRCBs is most influenced by beam depth (h), with the shear span-to-depth ratio (λ) and concrete compressive strength (fcl,150) being the subsequent key contributors. A comparative assessment between the XGBoost model and traditional analytical models was carried out to estimate the shear strength of CRCBs. Results demonstrate that the XGBoost model delivers enhanced predictive accuracy and improved performance. A parametric investigation examined its robustness under variations in geometry and material properties, while a user-friendly interface was created to support its practical use.