Influence of Graphene on Split Tensile Strength of Geopolymer Concrete: An Interpretable ML and PDP-Based Approach
Abstract
The need to curtail carbon emissions related to the production of cement has seen a growing interest in geopolymer concrete (GPC) as a viable and sustainable OPC alternative, and in the addition of graphene as a transformative nano-additive that improves tensile capacity. In this study, four machine learning models for predicting the split tensile strength (STS) of graphene-reinforced GPC were benchmarked using PDP analysis to interpret patterns in predictions in terms of: binder reactivity, nanoreinforcement mechanisms, and curing effects. The models were built on a dataset of 400 experimental measurements with seven input parameters, and their performance was evaluated using R2, RMSE, MAE, and MAPE. The DT model attained the highest accuracy, with training and testing R 2 of 0.97 and 0.94, RMSE of 0.21–0.31 MPa, MAE of 0.14–0.21 MPa, and MAPE of 0.051–0.109. Whereas RF demonstrated good generalization (R 2 = 0.94), ANN showed reasonable accuracy (R 2 = 0.91-0.89) with overfitting issues, and LR recorded the worst performance (R 2 = 0.88-0.85), which underscores its inability to capture the nonlinear binder-strength relationships that govern GPC systems. The most significant parameters identified through PDP analysis included GGBS (Δ = 5.3 MPa), curing temperature (Δ = 3.6 MPa), and graphene content (Δ = 3.4 MPa) with optimal dosage of graphene at 0.06-0.08 percent binder weight confirming the presence of crack-bridging and matrix-densifying effects. The study underscores the value of tree-based models as interpretable and robust tools for AI-driven mix optimisation, advancing the design of low-carbon, nano-reinforced GPC for structural applications.