Using Machine Learning to Find Environmentally At-Risk Communities

Abstract

Environmental health persists as a genuine concern in many US localities. However, public agencies often face limited capacity and resources to collect comprehensive environmental health data. Inspired by CalEnviroScreen, an environmental health assessment tool used to identify environmentally at-risk communities in California, I calculate pollution burden scores at the census tract level for the entire contiguous United States. Pollution burden is a composite score that encompasses 12 environmental (air, water, waste) indicators. I combine the actual pollution burden indicator data with predicted statistics using machine learning. I create an interactive, publicly accessible National Pollution Burden Map using ArcGIS Online. Although applied to US states, the same approach can also be applied to other regions of the world.