Developing a novel strength predictive modeling for 3D printable geopolymer concrete: An interpretable data-driven approach

3D printed geopolymer concrete (3DP-GPC) offers the potential for faster construction and intricate design implementation. However, optimizing the mix design remains a major challenge due to the complexity of geopolymer chemistry and the reliance on trial-and-error experimentation. Prior studies using machine learning (ML) have largely focused on fiber-reinforced GPC, leaving plain 3DP-GPC underexplored despite its fundamental importance for material optimization in 3D printing. This study addresses this gap by systematically applying and comparing four gradient-boosting ML models, XGBoost, LightGBM, CatBoost, and NGBoost, to predict compressive strength (CS) and flexural strength (FS) of plain 3DP-GPC. A dataset of over 500 entries was compiled from literature, normalized in terms of oxide composition of activators and curing parameters to reduce heterogeneity. The dataset was divided into training (80 %) and testing (20 %) subsets, with hyperparameter tuning performed via 5-fold cross-validation using GridSearchCV from Scikit-learn to ensure robust and computationally efficient model training. The models achieved high accuracy in predicting the CS of 3D-GPC, with R² values ranging from 0.900 to 0.902 (training) and 0.899–0.900 (testing). FS prediction was also promising, with R² values of 0.864–0.877 (training) and 0.791–0.808 (testing). SHAP (Shapley Additive Explanations) analysis identified curing period, water-to-binder ratio, and activator modulus as the most influential parameters for CS, while binder type and curing conditions dominate FS. Overall, this work provides systematic and interpretable ML framework for plain 3DP-GPC. The findings offer practical insights into key governing parameters and present a data-driven tool to streamline mix design, reducing experimental workload and accelerating the adoption of sustainable 3D printing in construction.