Managing waste for production of low-carbon concrete mix using uncertainty-aware machine learning model.

This study introduces an uncertainty-aware AI-driven optimization framework for designing sustainable concrete mixtures that incorporate waste-derived materials. The primary objectives are to reduce global warming potential (GWP) and promote a circular economy within the construction sector while preserving mechanical characteristics. A comprehensive dataset comprising 3114 unique concrete mix designs was developed from peer-reviewed literature, encompassing a wide range of parameters including cement (102-783 kg/m), water (80-247 kg/m), various supplementary cementitious materials (SCMs) (0-715 kg/m), waste-derived fine aggregates (0-740 kg/m), coarse aggregates (0-1095 kg/m), natural coarse (0-1380 kg/m), and fine aggregates (0-992 kg/m). The Explainable Boosting Regressor (EBR) demonstrated the highest predictive accuracy, with R values of 0.91 for compressive strength and 0.96 for GWP. The Weighted Jackknife + method provides narrow, localized uncertainty bounds, thereby enhancing the model reliability for heterogeneous mixes. Analyses of feature importance and shape functions identified cement, curing age, water, and GGBS as the most influential parameters. High-strength mixes are associated with a greater environmental impact owing to increased cement usage. A multi-objective optimization approach was employed, utilizing EBR as the objective function for compressive strength and GWP, whereas polynomial regression was used for the cost objective function to minimize the GWP and material cost while satisfying user-defined strength constraints. The framework incorporated nonlinear constraints in accordance with IS:10262-2019 standards. A graphical user interface (GUI) was developed for practical deployment. Among the evaluated combinations, the ternary blend of cement, GGBS, and waste glass achieved the highest waste utilization (960 kg/m), whereas fly ash and metakaolin-based mixes demonstrated a greater avoided GWP while meeting the M35-grade requirements. These findings underscore the potential of interpretable uncertainty-aware AI tools for guiding sustainable concrete design at scale.