Magnesium oxychloride (MOC) cement offers a sustainable alternative to Portland cement with superior mechanical properties, yet its practical adoption is hindered by poor water resistance. Various supplementary cementitious materials (SCMs) and chemical additives can improve strength retention (SR) after exposure to water. However, MOC cement design to attain desirable compressive strength (CS) and SR requires a precise selection of input materials, mix proportions, and additives, which is quite challenging. Therefore, employing machine learning techniques could facilitate and accelerate the development of MOC binders with the desired properties. To address this, we employ an interpretable machine learning (ML) framework leveraging four algorithms—Support Vector Regression (SVR), Random Forest (RF), Gradient Boosting (GBR), and XGBoost (XGB)—to predict CS and SR using two literature-derived databases (289 CS / 184 SR datapoints). GBR and XGB emerge as optimal predictors, demonstrating exceptional accuracy (testing R² ≥ 0.90) for both properties. Shapley Additive Explanations (SHAP) analysis reveals critical design rules: CS is driven by age, additive dosage (AD), SCMs, and molar ratios (M = MgO/MgCl₂•6H₂O; H = H₂O/MgCl₂•6H₂O), with age and M showing positive correlations while H, AD, and SCMs exhibit negative impacts. SR is governed by AD, water exposure duration (DEW), SCMs, and ratios (M, H), where AD, SCMs, and M show positive correlations with SR, while H and DEW exhibit negative impacts. SHAP-derived feature contributions and experimental observations from existing literature are consistent. The interpretable ML strategy used in this study will aid in the production and performance tuning of CS and SR of sustainable MOC cement for widespread applications.

Magnesium Oxychloride Cement, Water Resistance, Compressive Strength, Machine Learning Models, SHAP Analysis