Capillary water absorption values estimation of building stones by ensembled and hybrid SVR models.

The article focuses on predicting the capillary water absorption (CWA) values of building stones using four support vector regression (SVR) models: a conventional SVR, two ensembled SVR models (additive regression SVR and bagging SVR), and a hybrid SVR model optimized by the whale optimization algorithm (WOA-SVR). Using a dataset of diverse rock samples from Anatolia, Turkey, the models employed physical and mechanical properties—P-wave velocity (Vp), water absorption by weight (Wa), apparent porosity (n), and dry unit weight (ρd)—as inputs to forecast CWA. Results indicate all models performed reasonably well (R² > 0.79), with the hybrid WOA-SVR model achieving the highest accuracy (R² up to 0.913 for training and 0.886 for testing) and lowest error metrics, outperforming conventional and ensembled SVR models. Sensitivity analysis showed that dry unit weight and water absorption by weight were the most influential input variables for accurate CWA prediction. This study demonstrates that hybrid machine learning models can effectively estimate CWA, potentially reducing the need for costly and time-consuming experimental tests in assessing stone durability.