The U.S. Environmental Protection Agency (EPA), like other regulators, often relies on in-person facility inspections to identify violations of environmental regulations. However, these inspections are costly, and resources for regulatory enforcement are scarce, so only a small fraction of regulated facilities is inspected each year - raising the stakes for innovative, low-cost solutions to identify violators more efficiently.  

Since 2017, the Energy & Environment Lab has worked with EPA staff from the EPA headquarters, regional offices, and state environmental agencies to develop machine learning (ML) models to improve inspection targeting practices. Ultimately, these models promise to revolutionize the detection of environmental harm by utilizing data already at inspectors’ disposal to direct them to the facilities where their inspections will be the most likely identify violations and so improve compliance.  

The E&E Lab first developed a robust ML model harnessing administrative data spanning nearly two decades to inform inspection targeting for facilities that manage hazard waste under the Resource Conservation and Recovery Act (RCRA). The Lab and the EPA then conducted a randomized field evaluation to compare the model’s predictions to status quo inspection targeting practices. With promising preliminary results in hand, the EPA has made this ML targeting tool available to regulators nationwide. The E&E Lab is currently developing ML models to improve inspection targeting for facilities regulated under other federal laws such as the Clean Water Act and the Clean Air Act.