Framework reliability analysis on strength estimation of graphene nanoplatelets reinforced cementitious composites

Abstract Graphene nanoplatelets ( GrN ) can significantly improve the mechanical and functional performance of cement-based composites. Their high conductivity and large surface area also support the development of smart materials for structural health monitoring. However, traditional tree-based and fuzzy models struggle to represent the complex nanoscale interactions that influence the compressive strength of these composites. This study develops two optimized machine learning frameworks-ANFIS-ArP and DT-ArP-to predict the compressive strength of GrN -reinforced cementitious composites with high accuracy. A comprehensive dataset collected from published experimental studies was used to train the models. The Arctic Puffin Optimization (ArP) algorithm was implemented to tune the hyperparameters and enhance model reliability. The ANFIS-ArP model achieved the highest accuracy, with coefficient of determination values of 0.9848 during training and 0.9793 during testing. The DT-ArP model also performed well, with coefficient of determination values of 0.9715 and 0.9649. These results confirm the effectiveness of optimization-assisted machine learning for capturing nonlinear behavior in nano-modified composites. The study provides an efficient predictive tool for material designers and supports the development of durable and high-performance cementitious systems enhanced with graphene nanoplatelets.