Explainable machine learning models for corrosion degree prediction based on corrosion-induced crack width

The width of corrosion-induced cracks is a useful indicator for evaluating the corrosion degree of steel reinforcements. However, the existing empirical formulas do not account for the various factors that affect the corrosion cracking process. In this study, the corrosion degree prediction model based on the width of corrosion cracks was established using machine learning (ML) techniques. A total of 3,579 samples were collected under various experimental conditions from 56 published studies. Using these data, three ensemble machine learning (EML) models, namely random forest (RF), extreme gradient boosting machine (XGBoost), and light gradient boosting machine (LGBM), were developed. Among these, the RF model achieved the highest performance on the test set with a R² value of 0.84. In addition, SHapley Additive exPlanations (SHAP) was applied to interpret the data-driven model. A strong linear relationship between crack width and the corrosion degree was observed. The analysis also revealed that crack width, rebar diameter, and current density were the three most influential factors on the predict results. The SHAP interpretation results were consistent with theoretical understanding, demonstrating that the model effectively reflects the influencing mechanisms of multiple factors.