Using machine learning to make government spending greener

Abstract

In the face of the Triple Planetary Crisis concerning climate change, biodiversity loss, and pollution, the global community is in dire need of quantitative, data-based approaches to inform its response and guide its path towards a sustainable and equitable future. Government spending and fiscal policy are key levers in shaping this response. In order to assess the potential for using machine learning to inform policymakers' and governments' decision-making and spending allocation decisions based on environmental outcomes, the United Nations Environment Programme (UNEP) and the United Nations Conference on Trade and Development (UNCTAD) collaborated to produce a joint pilot study. The study uses official development assistance data (ODA) to train machine learning models to predict deforestation rates in six different countries: the Democratic Republic of the Congo, Haiti, Liberia, Madagascar, Solomon Islands, and Zambia. Initial modelling results were promising and the approach could prove to be a valuable asset to policymakers by enabling scenario analysis, where hypothetical budgets or spending allocations can be run through models trained on historical data to give insight on potential impacts on environmental indicators. Future research could be expanded to a pilot study with a national government using disaggregated budget data instead of ODA as model inputs.