Research on data driven and intrinsically explainable machine learning enabled compressive strength prediction and intelligent design of smart cementitious composites

This study addresses the problems that evolution of compressive strength in smart cementitious composites (SCCs) becomes complex due to the nanofillers and the low efficiency of optimal mix design based on repetitive testing. A data-driven framework for intelligent prediction and explainable analysis is developed. A compressive strength database is built covering four nanofillers: multi walled carbon nanotubes, graphene nanoplatelets, graphene oxide, and nano carbon black. The framework uses an explainable boosting machine (EBM) to learn the nonlinear mapping between eleven input features and compressive strength, enabling accurate prediction of SCC compressive strength and comparison with five mainstream machine learning models. The results show that EBM achieves highest predictive accuracy on the test set. Interpretability analysis identifies water-binder ratio, nano filler-cement ratio, and fine aggregate content as key factors governing the predicted compressive strength. These findings provide theoretical and technical support for rapid, performance-based intelligent mix design of SCCs.