High performance fine-grained geopolymer concrete using fly ash and ground granulated blast furnace slag: experimental results and modelling

This study investigates high-performance fine-grained geopolymer concrete (HPFGGC) produced from fly ash (FA), ground granulated blast-furnace slag (GGBFS), and solid sodium metasilicate (SMS), and evaluates Random Forest (RF), XGBoost, and LightGBM models for compressive-strength prediction. Under water curing at 27 ± 2 °C, the highest 28-day compressive strength reached 60.21 MPa. Increasing the GGBFS fraction generally improved compressive strength, although the mixtures achieving maximum compressive and flexural strengths differed. The dataset comprised 240 measurements from 10 mixture conditions and 40 mixture–age combinations. The predictors were GGBFS/(FA + GGBFS), water-to-binder ratio (W/B), and curing age. Model performance was evaluated using 10 repetitions of W/B-balanced five-fold validation grouped by mixture condition, with hyperparameter tuning performed within grouped inner folds. Mean outer-test R2 values were 0.7029, 0.7743, and 0.7789 for RF, XGBoost, and LightGBM, respectively, with corresponding RMSE values of 4.8928, 4.2178, and 4.1706 MPa. XGBoost and LightGBM showed comparable predictive performance. Permutation importance consistently ranked curing age and GGBFS fraction above W/B across all three models. These findings demonstrate the potential of ensemble-learning models for predicting HPFGGC compressive strength within the investigated experimental domain.