An Intelligent Framework for Sustainable Environmental Impact Assessment Using AI and ML

Abstract

Rapid environmental screening of construction materials is constrained by heterogeneous Environmental Product Declarations (EPDs), incomplete early-design information, and the time required for conventional life-cycle assessment (LCA). This study proposes an explainable artificial intelligence and machine-learning framework that predicts four cradle to-gate indicators: embodied energy, global warming potential (GWP), freshwater use, and construction-product waste. A reproducible 720-record synthetic benchmark was generated for six civil-engineering material classes using physically plausible ranges calibrated to published EPD/LCA literature; it is explicitly a computational benchmark rather than a field measured or manufacturer-certified database. Seven design and supply-chain attributes were used as predictors. Linear regression, support vector regression, random forest, gradient boosting, and Extra Trees models were trained on 80% of the records and independently evaluated on 20%. The best test performance reached R² values of 0.975 for embodied energy, 0.990 for GWP, 0.947 for water use, and 0.970 for waste. Permutation analysis identified material class, renewable energy share, binder content, recycled content, and transport distance as the dominant variables. The framework converts predicted indicators into a normalized Sustainable Environmental Impact Score and supports early comparison of material alternatives. It is intended as a rapid screening and decision-support layer, not as a substitute for standards-compliant project LCA. The findings demonstrate that harmonized data and interpretable ensemble learning can accelerate low-impact material selection while preserving traceability and uncertainty awareness.