Machine learning (ML) algorithms are increasingly used to model and predict economic outcomes. Using 15 years of data and nearly 10,000 variables, we build an ML model to predict the likelihood that manufacturing facilities will violate EPA regulations on hazardous waste. Given that the EPA can inspect a limited number of these facilities per year, we simulate the case in which the EPA’s inspection choices are replaced by facilities predicted to be high risk by our model. The results suggest that our model’s predictions improve on the EPA’s rate of finding violations by 50%. To validate our estimates of the model’s efficacy at improving targeting, we run a multi-year field test in which the EPA and the model each choose half of the facilities to be inspected. A field test that incorporates real world implementation challenges is critical for agency adoption. Ours is the first direct test of potential for machine learning to improve on the decision based targeting of government resources in the U.S.
Speaker: Katherine Meckel, UC San Diego
Contact:Website: Click to Visit
Save this Event:iCalendar
Windows Live Calendar