Data‑driven prediction of masonry flexural bond strength: Interpretability and implications

Flexural bond strength is a key parameter influencing the structural performance of masonry, yet its accurate prediction remains challenging due to the complex interplay of multiple material and testing variables. This study presents a machine learning (ML) framework for predicting flexural bond strength using a harmonized database comprising 1041 test specimens. Additionally, a review of 67 published studies was conducted to contextualize key influencing factors and inform the selection of input variables. Five baseline ML models were selected to capture a wide range of learning paradigms: artificial neural network (ANN), generalized additive model (GAM), random forest (RF), support vector machine (SVM), and eXtreme gradient boosting (XGBoost). These models were integrated using a stacking ensemble approach, with XGBoost as the meta-learner. Model performance was assessed using the coefficient of determination (R²), root mean square error (RMSE), and mean absolute error (MAE). Both XGBoost and RF yielded identical prediction errors (RMSE = 0.18 MPa, MAE = 0.12 MPa), though XGBoost achieved a slightly higher R² value (0.75 vs. 0.73), indicating a modest gain in explained variance. The stacking ensemble achieved the highest overall accuracy, with an R² value of 0.81, and a reduced prediction error (RMSE = 0.16 MPa and MAE = 0.11 MPa). To improve model interpretability and practical relevance, feature importance analysis and partial dependence plots (PDPs) were used to identify and visualize the effects of key predictors. Mortar compressive strength was found to be the most influential variable, followed by 24-hour water absorption, unit compressive strength, and testing standard. This study presents an interpretable ensemble-based prediction framework for flexural bond strength, supporting material selection, structural design, and sustainable masonry construction.