Machine learning models for predicting the compressive strength of sustainable recycled aggregate concrete incorporating supplementary cementitious materials

Abstract

The incorporation of supplementary cementitious materials (SCMs) and recycled aggregates (RAs) represents a pivotal strategy for mitigating the environmental impact and resource consumption of concrete production. To advance the utilization of such low-carbon concrete, this study develops a machine-learning framework for predicting the compressive strength of concrete incorporating fly ash and ground granulated blast furnace slag. A dataset of 230 mixtures was compiled from peer-reviewed studies retrieved via Google Scholar, the Web of Science and the China National Knowledge Infrastructure. Keywords for the search included “Recycled aggregate concrete”, “Fly ash”, “Ground granulated blast furnace slag”, “Compressive strength”, and “Machine learning”. Studies were included if they reported compressive strength and curing age, and provided sufficient mix-design information for feature extraction. Incomplete, duplicate, overlapping, or non-comparable records were excluded. The resulting dataset from the literature was analyzed using seven key mix parameters as inputs. Distinct from prior RAC machine learning studies, this work targets low-carbon recycled aggregate concrete incorporating both fly ash and ground granulated blast furnace slag, and systematically benchmarks conventional, ensemble, and optimization-enhanced neural models, complemented by sensitivity interpretation and a deployable GUI tool. Six machine-learning methods were evaluated, including conventional models and optimization-enhanced neural networks. The particle swarm optimization-based backpropagation (PSO-BP) model achieved the best predictive performance, attaining a determination coefficien@@t (R Chithra et al. (2016)2) of 0.953 and a root mean square error (RMSE) of 3.39 MPa, alongside demonstrating superior generalization capability compared to single-algorithm models. Curing age and cement content were identified as the dominant factors governing strength development. Furthermore, a user-friendly prediction tool was developed to support rapid strength estimation and reduce laboratory testing efforts. This framework establishes a practical and reliable pathway for optimizing the design of low-carbon recycled aggregate concrete, thereby facilitating its broader application in sustainable construction.