Machine-learning-based prediction of the compressive strength of superabsorbent polymer-incorporated cementitious materials: Modeling and experimental insights into polymer characteristics

To predict the compressive strength of superabsorbent polymer (SAP)-incorporated cementitious materials, this study developed machine-learning-based models incorporating SAP characteristics (particle size, shape, type, and absorption capacity) and evaluated them in tandem with using X-ray micro-computed tomography (CT) and compressive strength testing of the resulting cementitious material. CatBoost provided the best performance (coefficient of determination = 0.9680 and accuracy = 92.70%) owing to its effective handling of nonlinearity and feature interactions, as well as its robust treatment of categorical variables through ordered boosting and target statistics. The relative importance of the SAP-related input features affecting compressive strength prediction followed the order of absorption capacity, dosage amount, type, particle size, and shape. Experimental observations combining X-ray micro-CT measurements and compressive strength testing confirm that in addition to total porosity, variations in pore distribution, shape, and hydration degree induced by SAPs contributed to the compressive strength of the cementitious material. • CatBoost predicted strength well for SAP-incorporated cementitious materials • Superabsorbent polymer (SAP) parameters are key predictors of compressive strength • Experiment unveiled SAP effects on concrete pore structure and compressive strength