Applying machine learning to a U.S. Environmental Protection Agency initiative, researchers reveal how key design elements determine what communities bear the burden of pollution. The approach could help ensure fairness and accountability in machine learning used by government regulators.