Machine learning‐based prediction and physicochemical interpretation of compressive strength development in metakaolin‐based geopolymer concrete

Abstract

The demand for sustainable construction materials has spurred the development of alkali‐activated geopolymer concrete as a low‐carbon alternative to Portland cement. Fly ash–GGBFS blends enhance mechanical properties while reducing environmental impact. Predicting the compressive strength of these systems is complex and depends on mixture composition, activator, and curing. This study analyzes a final, validated database of 226 metakaolin‐based geopolymer concrete mixtures using six predictive approaches: the Non‐Linear model (NL), Power model (P), Quadratic Model 1 (Q1), Quadratic Model 2 (Q2), Linear–Inverse Mixed model (LIM), and Artificial Neural Network (ANN) to assess parameter influences on strength. Model performance was evaluated using R ² , adjusted R ² , RMSE, MAE, MAPE, VAF, and RSE. Sensitivity analysis was also conducted to assess the influence of Na ₂ O/Al ₂ O ₃ , SiO ₂ /Al ₂ O ₃ , and H ₂ O/Na ₂ O ratios on strength development. The ANN achieved the highest predictive accuracy. The findings provide mechanistic insight and a practical, data‐driven framework for optimizing sustainable MK‐GPC mixtures.