Abstract This study investigates the combined influence of binder substitution and alkaline activator dosage on the mechanical performance of alkali-activated materials incorporating ground granulated blast-furnace slag using an integrated experimental and machine-learning-based framework. A comprehensive dataset comprising mix-design parameters and compressive-strength results under different curing conditions was analysed using multiple predictive models. The developed models demonstrated strong agreement between measured and predicted strengths, indicating robust predictive capability across the investigated parameter space. Ensemble tree-based models demonstrated superior predictive capability, achieving coefficients of determination ( R 2 ) of 0.87–0.89 on the testing phase with root mean square error between 10.5 and 11.2 MPa and mean absolute errors generally below 10 MPa. Feature-importance and interpretability analyses, including partial dependence plots, revealed the relative contribution of key variables, identifying binder composition and activator dosage as dominant factors governing strength development, while curing conditions exhibited secondary but non-negligible effects. The results provide quantitative insight into how variations in binder substitution levels and activator content influence material performance, enhancing understanding of parameter sensitivity and interaction effects. Overall, the findings support data-driven assessment of mix-design strategies and highlight the potential of machine-learning tools for interpreting performance trends in alkali-activated materials.
Al-Naghi et al. (Thu,) studied this question.