Machine learning-based estimation of CO2 footprint and environmental-mechanical performance of blended cement concrete

Blended cements with supplementary cementitious materials (SCMs) are extensively used to mitigate the environmental impact of concrete. However, assessing their environmental performance often requires detailed data and time-intensive analyses. This study presents a machine learning-based methodology for the rapid estimation of the CO₂ footprint and environmental-mechanical performance (CO₂/MPa ratio) of concrete mixtures using only the proportions of their main components (i.e., ordinary portland cement (OPC), SCMs, aggregates, water, and water-reducing admixtures). The models were developed using a dataset of 246 mixtures compiled from the literature and validated against 15 experimentally tested mixtures. The results demonstrate that the Gaussian Process Regressor provides the highest predictive accuracy for both CO₂ footprint and CO₂/MPa ratio. Feature analysis revealed that OPC content has the highest impact on the CO₂ footprint, while aggregate fraction has the most significant influence on the CO₂/MPa ratio. An optimization framework was also implemented to explore the trade-offs among mix components, showing that increasing SCM content does not always lead to improved CO₂/MPa ratio, highlighting the need for balanced mixture design. The developed models offer a practical tool for supporting early-stage decision-making in construction projects by enabling rapid sustainability assessments of concrete mixtures, independent of specific SCM types, production methods, or geographical context. • GPR model best predicts CO₂ footprint and CO₂/MPa ratio of concrete. • OPC fraction has the highest impact on the CO₂ footprint of concrete. • Aggregate fraction has the highest impact on the CO₂/MPa ratio of concrete. • Optimal OPC–SCM mixes minimize CO₂/MPa, not always with more SCM replacement. • GPR models estimate concrete’s environmental impact before material selection.