Interpretable Machine Learning Framework for Strength and Waterproofing Design of Styrene-Acrylate Mini-emulsion Modified Cement Mortars

Cement-based repair mortars and surface layers are prone to cracking and water ingress, which accelerate durability loss and increase maintenance demands in building and infrastructure systems. This study investigates styrene–acrylate mini–emulsion (StAcEm)-modified cement mortars and develops an integrated experimental hybrid machine-learning (ML) framework to predict and optimize their mechanical and waterproofing performance. A StAcEm was synthesized by copolymerizing styrene (St), butyl acrylate (BA), 2-ethylhexyl acrylate (EHA), and methyl methacrylate (MMA), and characterized via average particle size (Avg PS), polydispersity index (PDI), zeta potential, water contact angle (WCA), and hard monomer-to-soft monomer (HM/SM) ratio. Mortars were prepared over practical ranges of polymer-to-cement (P/C) and water-to-cement (W/C) ratios and curing ages (3d, 7d, 28d) and tested for compressive strength (CS), flexural strength (FS), water absorption (WA), and water penetration resistance (WPR). Five ensemble regressors (CatBoost, Random Forest, XGBoost, LightGBM, Bagging) were trained; the best CatBoost model achieved the test-set R 2 of 0.938 (RMSE = 0.450 MPa) for FS and 0.961 (RMSE = 2.008 MPa) for CS, indicating high accuracy and robust generalization. SHAP-based analyses showed that CS and FS, together with P/C, W/C, WA, and WPR, are the dominant predictors, with thresholds where CS > 16–18 MPa promotes FS, while P/C > 1.8–2.0%, WA > 5%, and WPR > 7–8 mm reduce predicted strength. The novelty of this work lies in an experimentally grounded, interpretable ML framework that integrates StAcEm physicochemistry, mixture design, curing age, and engineering performance data to quantify polymer effects and jointly optimize the mechanical and waterproofing properties of durable StAcEm-modified cement mortars.