This is the authors' abstract. We don't add key points for this paper.
The construction industry requires sustainable alternatives to ordinary Portland cement, but predicting the environmental impact of geopolymer concretes remains complex due to highly variable mixture designs. This study aims to develop a robust machine learning framework to accurately predict the volumetric CO 2 -eq footprint of ground granulated blast furnace slag geopolymer concretes and interpret the underlying chemical drivers. A comprehensive database comprising 139 experimental mixture designs and curing conditions was systematically compiled from peer-reviewed literature. Subsequently, the corresponding environmental outputs were not measured experimentally, but rather calculated using a deterministic lifecycle assessment framework based on established emission factors. Eight distinct machine learning architectures, including decision trees, neural networks, and support vector regression, were trained and evaluated using a rigorous 5-fold cross-validation and grid search methodology. Performance evaluation revealed that the support vector regression model achieved superior predictive accuracy and generalization, with a testing R 2 of 0.951and a mean squared error of 78.9, effectively mitigating the severe overfitting issues observed in the tree-based and neural network models. To resolve the black-box nature of the predictive algorithms, Shapley Additive Explanations were integrated to quantify feature importance and interaction dynamics. The interpretability analysis identified superplasticizer dosage and dry sodium hydroxide content as the predominant drivers of CO 2 -eq emissions, reflecting the intensive embodied carbon of synthesized chemical additives compared to bulk aggregates. Ultimately, this study provides a highly accurate, interpretable computational tool that empowers engineers to predict geopolymer concrete formulations for assessing environmental impact while maintaining desired stoichiometric balances.
Answers are grounded in this paper and cite their sources.
No takes yet. Share an insight, caveat, or question.
Zhang et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: