Advanced explainable models for strength evaluation of self-compacting concrete modified with supplementary glass and marble powders

Abstract Self-compacting concrete (SCC) is increasingly adopted in modern construction due to its self-flowing nature, which eliminates the need for mechanical vibration and enhances construction quality. The use of industrial waste materials like marble powder (MP) and glass powder (GP) in SCC presents a sustainable alternative to conventional materials, reducing environmental impact. However, predicting the compressive strength (CS) of such mixes through traditional testing methods is time-consuming, costly, and limits rapid mix optimization. This motivates the adoption of machine learning (ML) techniques, which can efficiently analyze complex datasets and identify patterns that influence concrete performance. In this study, three ML models, gradient boosting, bagging regression, and random forest (RF), were used to predict the CS of SCC incorporating MP and GP. Among them, RF achieved the highest accuracy ( R ² = 0.95). Model interpretability was ensured through Shapley Additive exPlanations, partial dependence plots, and individual conditional expectation analyses, which identified curing time as the most influential feature. The Taylor plot and validation metrics confirmed RF’s superior reliability. This research highlights the potential of ML not only as a predictive tool but also as a means of understanding key factors in sustainable mix design, ultimately promoting smarter and greener construction practices.