Data-informed multi-scale prediction of compressive strength in concrete with ground recycled cement

This study presents a comprehensive, data-driven investigation into the compressive strength (f’c) of concrete incorporating ground recycled concrete cement (GRC), an emerging sustainable binder. A dataset of over 900 concrete and mortar mixtures compiled from the literature was analyzed, encompassing mix design parameters, physical and chemical properties, and curing conditions. Seven machine learning algorithms were systematically evaluated across three progressively enriched feature sets. Model evaluation primarily used a random 80/20 training-test split, with hyperparameter tuning performed via standard 5-fold cross-validation within the training set. As an additional robustness check, publication-wise group-based cross-validation was conducted, ensuring that all mixtures from the same study were kept together in either training or validation folds, thereby assessing cross-study generalization. XGBoost consistently achieved the highest predictive accuracy (R2 > 0.958), outperforming other models, while CatBoost performed comparably under limited descriptor scenarios. The inclusion of detailed binder-specific descriptors, such as particle size, oxide composition, and supplementary cementitious material proportions, significantly enhanced predictive performance, highlighting the value of multi-scale material characterization in capturing the heterogeneous behavior of GRC-based concretes. Explainable AI via SHAP analysis identified curing age, GRC content, and water-to-binder ratio as dominant contributors to model predictions, while microstructural descriptors exhibited increased attribution in feature-enriched scenarios; these findings reflect model-based associations rather than direct mechanistic causation. These results provide a rigorous, interpretable framework for optimizing sustainable concrete mixtures and highlight the potential of integrating materials science with machine learning to accelerate the design of next-generation construction materials.