Influence of chemical components and molar ratios on strength development of one-part alkali-activated mortar: Ensemble machine learning models

One-part alkali-activated mortar (OPAAM) is an emerging innovation in construction materials, establishing itself as a sustainable alternative to Portland cement mortar. Numerous studies examined how CaO content, water-to-binder (w/b) ratio, and molar ratios (SiO 2 /Na 2 O, SiO 2 /Al 2 O 3 , Na 2 O/Al 2 O 3) influence OPAAM’s compressive strength. However, their combined influence in a unified and interpretable framework has not been investigated, which limits practical mix design decisions and encourages trial-and-error experimentation. This study introduces ensemble machine learning (ML), specifically stacked models integrated with SHAP (SHapley Additive exPlanations), to provide transparent insight into how physical (e.g., w/b ratio, binder and solid activator content) and chemical (such as CaO percentage and molar ratios) parameters collectively affect strength, enabling data-driven screening of promising OPAAM mixtures before laboratory confirmation. A dataset of 141 samples was used which include eight input features and 28-day compressive strength as the target. After cleaning and standardization, models were trained using 5-fold cross-validation with grid-search hyperparameter tuning. The best configuration combined decision trees, gradient boosting, and random forests as base predictive models, with CatBoost as a meta-model, achieving determination coefficient (R 2 = 0.8824) and root mean square error (RMSE = 5.56 MPa) for the modeling testing phase. SHAP, using a surrogate CatBoost trained on original features, identified (SiO 2 /Na 2 O, SiO 2 /Al 2 O 3 , Na 2 O/Al 2 O 3 and CaO as the most influential variables. These insights align with prior experiments. By pairing high-accuracy stacking with explainable feature attributions, the framework provides a robust and interpretable tool for OPAAM strength prediction. It accelerates mix design optimization and reduces early-stage experimental burden by enabling data-driven screening before laboratory confirmation. It also supports the development of high-performance, eco-friendly OPAAM for broader construction applications.