Predictive modeling and optimization of eco-friendly cementitious grouts containing waste marble dust for semi-flexible pavements

Semi-flexible pavements integrate the advantages of flexible and rigid systems through a porous asphalt layer filled with cement-based grout; however, their high cement consumption raises environmental concerns. This study aims to optimize eco-friendly cementitious grouts incorporating waste marble dust (WMD) using machine learning techniques based on 72 experimental mixtures. The dataset included water–cement ratios (0.35–0.50), WMD content (0–25%), grout flow, and compressive strengths at 7 (CS_7d) and 28 days (CS_28d). Five machine learning models, including artificial neural networks (ANN), decision tree (DT), random forest (RF), support vector machine (SVM), and k-nearest neighbor (KNN) were developed and evaluated using mean absolute error (MAE), coefficient of determination (R 2 ), and adjusted R 2 . Among these, the decision tree model demonstrated the highest predictive accuracy with R² = 0.994 and MAE = 0.266 MPa. Explainable artificial intelligence methods, including SHapley Additive exPlanations (SHAP) and partial dependence plots (PDPs), identified 7-day compressive strength as the most influential parameter, as it provides an experimentally measured representation of the early-age strength development of the grout. Results indicate that increasing water–cement ratio and marble dust content reduces CS_28d. Model-based optimization allowed up to 32% cement replacement at a 0.4 water–cement ratio with 0.5% superplasticizer, while higher ratios reduced replacement efficiency. The findings demonstrate the potential of WMD as a partial cement replacement in semi-flexible pavement grout, supporting waste valorization and reducing cement demand while providing a data-driven approach for sustainable mixture design.