Predicting Broadband Expansion: An Analysis of Geographic and Demographic Influences

Abstract

High-speed Internet access (’broadband’) is available nearly everywhere in urban and suburban parts of the contiguous U.S. However, there are many rural areas which do not have broadband access yet. We implemented a machine learning framework consisting of multiple supervised models for accurately predicting broadband expansion using geographic and demographic data on a census block group level. We then utilized these models to gain an understanding of how these features relate to broadband expansion. Also, we discussed how this data must be considered differently when investigating unsubsidized broadband expansion rather than government subsidized expansion. Understanding relationships between geographic and demographic features with respect to broadband speed and accessibility is reliant upon publicly available data provided by the Federal Communications Commission (FCC), United States Census Bureau, and the Universal Service Administrative Company. The project included identifying, cleaning, and aggregating the data as well as creating a cloud-based-storage infrastructure with Google BigQuery. We found that by considering geographic and demographic data, effective models can be constructed to both understand previous broadband expansion and predict future expansion. Notably, we found that road, housing and population density, winter temperature, and elevation are of significant importance in determining whether an area receives broadband.