Prediction of the sustainable concrete strength using machine Learning-based boosting and expression techniques

This study investigates the feasibility of rice husk ash (RHA) in concrete and aims to predict its compressive strength (fc′), split tensile strength (fsp) and flexural strength (fst) using advanced machine learning methods, including Categorical Boosting (CatBoost), eXtreme Gradient Boosting (XGBoost) and Multi-Expression Programming (MEP). The data were sourced from previously published studies. ShaPley Additive exPlanations (SHAP) were applied to the XGBoost, while sensitivity and parametric analyses were performed for the MEP model to examine the influence of input variables on strength development. The proposed models were evaluated using statistical checks and external validation criteria. Results demonstrated that CatBoost and XGBoost exhibit higher predictive efficiency than the MEP model. However, the streamlined expressions generated by MEP remain suitable for practical design applications and are capable of accurately predicting fc′, fsp and fst of RHA concrete. In RHA concrete, the water–cement ratio and Age content greatly affect fc′. The most critical parameters for fsp and fst are (Age and Water) and (Cement and Water), respectively. This study provides insights for the design and optimisation of RHA concrete mixtures and underscores the potential of machine learning and statistical approaches in predicting and optimising the performance of construction materials.