Machine learning for low-carbon concrete strength prediction: methods and trends

Low-carbon concrete (LCC) is an effective approach to reduce carbon dioxide emissions in construction while maintaining adequate compressive strength (CS). Machine learning (ML) methods have been increasingly applied to predict the CS of LCC; however, existing studies remain fragmented in terms of data sources, material systems, modelling strategies, and evaluation criteria. This review provides a systematic synthesis of 172 peer-reviewed studies on ML-based CS prediction of LCC. Bibliometric analysis is conducted to evaluate the representativeness of the literature database, followed by an examination of research evolution across methodological and material dimensions. Mainstream ML approaches are analysed with respect to modelling mechanisms, feature engineering, and performance evaluation, and qualitatively compared using a unified five-point scoring framework. The results reveal a clear shift from single-model approaches towards ensemble learning, Bayesian optimisation, and multi-objective modelling, particularly for systems incorporating supplementary cementitious materials and recycled constituents. Key challenges and future research directions are discussed, including interpretable ML, climate-adaptive modelling, and emerging artificial intelligence-assisted design frameworks, to support more robust and generalisable ML applications in LCC design.