Innovative development of an advanced machine learning modeling for predicting compressive strength of graphene-infused concrete: a SHAP and PDP approach

Abstract

The quasi-brittle response of cementitious materials continues to limit their structural efficiency, necessitating advanced reinforcement strategies. While conventional fibers improve mechanical performance, the emergence of graphene-based nanomaterials offers new opportunities to tailor interfacial interactions and enhance matrix cohesion at multiple scales. This study presents a data-driven framework for predicting the compressive strength of graphene-modified concrete by integrating mix design parameters and machine learning techniques. A curated database of 371 mix compositions was assembled from published studies, incorporating seven mix variables along with the curing period. Three ensemble learning models, Extreme Gradient Boosting, Adaptive Boosting, and Bagging Regressor, together with two hybrid configurations, were developed and evaluated. Model performance was assessed using standard statistical indices, with particular emphasis on generalization across unseen data. Based on the outcomes of this study, the hybrid Extreme Gradient Boosting–Bagging model demonstrated the highest predictive capability, achieving R 2 values of 0.92 and 0.87 for training and testing datasets, respectively, outperforming individual and alternative ensemble models. Interpretability analysis using SHapley Additive Explanations identified water (+ 14.04) and curing age (+ 6.46) content as dominant contributors to strength development, while the influence of graphene and supplementary materials exhibited nonlinear and interaction-dependent effects. Furthermore, PDP analysis summarized that an increase in graphene content within the range of 0–2 kg/m 3 led to a rise in predicted strength from approximately 53 MPa to around 60 MPa. The proposed approach provides both high predictive accuracy and enhanced interpretability, offering a reliable tool for mix design optimization of graphene-enhanced concrete. More broadly, the study underscores the potential of hybrid machine learning frameworks to accelerate the development of next-generation cementitious composites with improved performance and sustainability.