Predicting compressive strength of red mud-based geopolymers via machine learning: Insights from experimental validation

Red mud, a challenging industrial by-product, poses a significant environmental challenge due to its increasing stockpiling. Its use as environmentally functional materials received increasing attention but requires good understanding of the compressive strength. This study employed machine learning (ML) approaches to develop a predictive model linking the composition (precursor, alkali activator) and curing conditions of red mud-based geopolymers (RMG) to their compressive strength. Among the nine studied ML algorithms, XGB model yielded the highest accuracy, achieving a coefficient of determination (R 2 ) above 0.9 in both training and testing. To improve the explainability of ML models, Shapley Additive exPlanations (SHAP) was employed to interpret the relative influence of input features on compressive strength prediction. The analysis highlighted the critical impact of calcium oxide content (RM_C), curing time (CC_CT), and alkali activator modulus (AC_S/N). Experiments were also designed based on these insights to validate the model's behavior concerning these key factors, with an average prediction accuracy error within 30%. Consequently, the machine learning based approach enhances the development efficiency for RMG, providing a new paradigm for the efficient materialization of similar solid wastes for sustainable development. • Machine learning models were developed to predict the compressive strength of RMG • XGB model achieved R 2 > 0.9 in both training and testing • SHAP analysis identified key strength-governing factors to guide experimental design • Experiments demonstrated the reliability of XGB model • Results can support the development of eco-friendly building materials from red mud