Corporate Carbon Footprint : A Machine Learning Predictive Model for Unreported Data

We propose a model based on statistical learning techniques to predict unreported corporate greenhouse gas emissions, which generates considerably better results than existing approaches. The model uses one linear and one non-linear learners only, which reduces its complexity to the minimum required. An iterative approach to detecting and correcting data significantly improves the model predictions. Unlike mainstream approaches, which tend to construct one model for each industry, we propose to construct one single global model that uses industries as a factor. This addresses the problem of lack of breadth or lack of reported data in some sectors and generates practical results even for industries where other approaches have failed. We show results for scope 1 and scope 2. Adapting the framework to scope 3 corporate emissions will be the focus of a future publication