A robust hybrid machine learning framework for multidimensional ESG assessment in critical mineral supply chains: The case of cobalt mining

The global energy transition is generating substantial demand for critical minerals such as cobalt, thereby positioning Iran, with its considerable reserves, as a significant potential supplier. Cobalt extraction poses environmental, social, and governance challenges that threaten sustainable development and supply-chain security. This research presents a hybrid machine learning framework, esg_assessor, designed for a comprehensive evaluation of the ESG risks and opportunities linked to cobalt mining in Iran. The framework integrates expert-elicited knowledge with a non-linear Gradient boosting (XGBoost) ensemble, offering a nuanced evaluation that balances transparency and predictive accuracy. The assessment highlights significant risks, including hydrogeological stress, complex pollution issues, and systemic governance failures such as corruption and social unrest. Conversely, it underscores substantial opportunities in sustainable mining practices, job creation, and economic revitalization. Model explainability analyses using SHAP, LIME, and ICE indicate that environmental factors are the primary determinants of the ESG Score, contributing approximately 4 times more (by mean absolute SHAP) than the other two pillars (Social and Governance). At the feature level, risks in “Hydrogeological Stress and Water Depletion” and “Ecosystem Fragmentation and Biodiversity Loss” appear to offer notable impact (mean SHAP ≈ 0.0075 and 0.006, respectively), whereas social-related constraints “Loss of Local Livelihoods” show a smaller net risk contribution (mean SHAP ≈ 0.001). esg_assessor delivers a transferable, uncertainty-aware ESG assessment framework that quantitatively integrates expert judgment with machine learning to support site-level screening, permit conditioning, and scenario-based policy design for critical mineral supply chains.