Abstract

Scope 3 emissions are major contributor for emissions of many industries. Since these emissions are indirect/complex in nature, data availability is a major challenge in estimating them. Current methods mostly rely on high level economic transactions with less inclusion of product details and have shortcomings in handling lack of data. In this research, we firstly discussed the benefits/shortcomings of existing methodologies for estimating scope 3 emissions and proposed a Machine Learning (ML) based framework to overcome those shortcomings. In our approach we map the Central Product Classification (CPC) system with International Standard Industry Classification (ISIC) and North American Industry Classification System (NAICS) together to gain higher granularity in scope 3 emission estimation. For ML models implementation, we used the supply chain emission factors based on NAICS codes as targeted variable. Our best performing model, Gradient Boosting Decision Trees, achieved an R² score of 0.9108 and MSE of 0.034 while giving balanced importance to CPC and ISIC codes. Results suggest our ML framework, combined with integrated classifications, can enhance the granularity and predictive accuracy for scope 3 emission factors derived from monetary input-output databases.

This work is licensed under a Creative Commons Attribution 4.0 License.