Machine learning based optimization of fly ash content for improving geopolymer concrete compressive strength.

The increasing demand for sustainable construction materials has intensified interest in fly ash-based geopolymer concrete (FA-GC) as an alternative to traditional concrete. This study aims to predict the compressive strength (CS) of FA-GC and optimize its mix design for maximum performance. Several artificial intelligence (AI) models-Tabular Prior-Data Fitted Network (TabPFN), Histogram-based Gradient Boosting (HistGBoost), M5Prime, and Automatic Feature Interaction Learning (AutoInt)-were applied to predict CS, with hyperparameters tuned via Optuna. Data were split into 75% training and 25% testing. TabPFN achieved the highest accuracy during training (R² = 0.981) and testing, with the lowest RMSE (4.02) and sMAPE (8.20%). Uncertainty analysis also showed TabPFN had the lowest CI (0.697) and R-Factor (2.730), indicating superior stability and predictive reliability compared to other models. For mix design optimization, four metaheuristic algorithms-Harris Hawks Optimization (HHO), Grey Wolf Optimization (GWO), Lyrebird Optimization Algorithm (LOA), and Polar Bear Algorithm (PBA)-were employed. HHO yielded the highest CS at 61.56 MPa, outperforming the others. Sensitivity analysis using SHAP values revealed that Al₂O₃ (%) and SiO₂ (%) most strongly enhanced CS, followed by Coarse aggregate (kg/m³). Partial dependence plots confirmed these trends, while Fine aggregate (kg/m³) and Duration (hr) had less influence. This study highlights that combining advanced AI models with metaheuristic optimization, sensitivity analysis, and uncertainty evaluation can significantly improve the prediction, stability, and mix design of FA-GC, paving the way for more efficient and sustainable geopolymer concrete solutions.