Advanced hybrid machine learning models for estimating chloride penetration resistance of concrete structures for durability assessment: optimization and hyperparameter tuning

This study explored advanced hybrid machine learning (ML) techniques for estimating the non-steady-state migration coefficient (D_nssm) of concrete, a key indicator of chloride penetration resistance. Support vector regression (SVR) was integrated with four metaheuristic optimization techniques: grey wolf optimization (GWO), gorilla troops optimization (GTO), firefly algorithm (FFA), and particle swarm optimization (PSO) to improve predictive accuracy. Among these models, SVR-GTO exhibited the superior effectiveness, attaining the maximum R² of 0.97 and the lowest root mean square error (RMSE) of 0.93. The SVR-GWO model similarly demonstrated the robust predictive accuracy, with an R² of 0.92 and an RMSE of 1.55, whereas the SVR-PSO and SVR-FFA models recorded slightly lower R² of 0.91 and 0.89, with RMSE of 1.67 and 2.13, respectively. To enhance model transparency and interpretability, the study employs SHapley Additive exPlanations (SHAP), partial dependence plots, and individual conditional expectation plots, offering a comprehensive understanding of how predictors affect the predicted outcomes. SHAP model revealed the higher significance of water-to-binder ratio (W/B), migration test (MT) age, and total aggregate (TA) in predicting the D_nssm. An interactive graphical interface was created to estimate the D_nssm of concrete, allowing efficient model interaction and eliminating the need for physical experimentation.