Increasing the Efficiency of Climate and Air Quality Regulation through Machine Learning at the United States Environmental Protection Agency

Facility inspections represent a key enforcement practice across regulatory domains, but they are costly, infrequent, and detect only a small fraction of violations. The US Environmental Protection Agency (EPA) typically relies on inspection targeting practices that are decades old. Since 2015, the Energy & Environment Lab at UChicago (E&E Lab) has worked with the EPA to pioneer the application of machine learning (ML) to improve enforcement targeting. The researchers developed a model that predicts which facilities are most likely to violate hazardous waste regulations and then ran a field test that demonstrated the model yielded a significant 82% increase in detection of serious hazardous waste violators compared to the EPA’s status quo practices. The EPA scaled the model nationwide and is now eager to scale ML-driven enforcement targeting to mitigate air pollution. The researchers propose to develop a ML model to predict violations under the Clean Air Act, to be scaled as an inspection targeting tool for regulators nationwide. The consequent increase in violation detection is expected to bring about large reductions in fugitive emissions of greenhouse gases and other pollutants harmful to health.