Microstructural and mechanical characteristics evaluation of alkali-activated slag concrete and optimize with...

This investigation employed both classical and hybrid machine learning (ML) strategies to examine the impact of alkali-activated materials on the compressive strength (CS) of Fly ash-based alkali-activated concrete (FAAC). For this reason, four supervised ML models were used in this study to estimate compressive strength. For experimental validation, FAAC specimens were produced with varying slag proportions and subjected to curing at two specific thermal conditions (25 °C and 90 °C). Compressive strength tests were conducted to evaluate the effects of slag content and curing temperature on the mechanical properties of FAAC. The modeling approach incorporated variables such as fly ash, composite slag, NaOH, Na₂SiO₃, fine aggregates (FA), coarse aggregates (CA), water-to-binder ratio (W/B), curing period, thermal conditions, and age. A dataset consisting of 650 samples, collected from existing literature and experiments, was partitioned into categories for training and testing. The model's accuracy was assessed employing performance indicators including the mean absolute percentage error (MAPE), root mean square error (RMSE), coefficient of determination (R²), and mean absolute error (MAE). The hybrid XGB-DT model attained the highest precision (R² = 0.986, RMSE = 2.44 MPa). Additionally, SHAP (SHapley Additive ExPlanations) and partial dependence plot (PDP) analyses were applied to identify perfect ranges for critical parameters affecting compressive strength. The most suitable combination of various input factors for designing FAAC mixtures was found using response surface modeling (RSM). According to a microstructural study, the increase of slag content improved strength by producing a denser microstructure and stronger C-S-H linkages.