Prediction of Properties of Sustainable Concrete Incorporating Supplementary Cementitious Materials and Recycled Aggregates using Machine Learning- A Systematic Review

Abstract

As the construction industry strives to cut carbon emissions, there has been growing interest in the use of supplementary cementitious materials (SCM) and recycled aggregate concrete (RAC). However, the interaction between the different kinds of SCM, recycled aggregate quality, water-to-binder ratio, curing age and admixtures is extremely non-linear, which cannot be adequately addressed by conventional mix-design and prediction approaches. Data-driven modelling of these interactions and optimization of concrete mixtures can be done with the help of artificial intelligence (AI) and machine learning (ML). In this systematic review, the author created a comprehensive list of peer-reviewed studies in the field of prediction of mechanical and durability properties of SCM-RAC concrete, using the PRISMA 2020 framework, from 2020 to 2025. The studies employed sample sizes from as low as 100 to as high as 59,412 records and models such as artificial neural networks, support vector regression, Random Forest, XGBoost, LightGBM, hybrid optimization models, Bayesian models, and deep learning architectures. The shift from single to ensemble and hybrid approaches are evident in the synthesis, with ensemble and hybrid approaches yielding better performance in general. The R² value is typically greater than 0.90 for compressive strength and can reach 0.98–0.99 when predicting durability optimized. The methods of explainability (SHAP and PDP) reveal that the cement content, W/B, curing age and dosage of SCM are the most dominant predictors, whereas the use of recycled aggregate absorption and parent concrete strength is not so extensively used. When combined with multi-objective optimization and life cycle assessment, the combination of ML can reduce cement use by 25-32% while also reducing global warming potential by 40-54%. There are several research gaps: lack of open datasets, lack of a modelling approach for blended SCMs, lack of explainability in durability studies, and lack of practitioner-oriented decision-support tools. With this in mind, the review identifies some future priorities regarding the design practice of SCM-RACs that are robust, transparent, and sustainable.