Authors: Charles Solberg*, CAMBA HomeBase
Topics: Urban Geography, Applied Geography
Keywords: Eviction, Displacement, Homelessness Prevention, Urban Geography, Applied Geography, Machine Learning
Session Type: Paper
Start / End Time: 8:00 AM / 9:40 AM
Room: Roosevelt 0, Marriott, Exhibition Level
Presentation File: No File Uploaded
In 2017, roughly 60,000 individuals slept in a New York City homeless shelter each night, an increase of nearly 82% between 2007 and 2017. While the causes of homelessness are variable and complex, eviction remains one of the most common reasons for shelter entry—especially among families with children—in New York City. Prior research has focused on effectively targeting and predicting a family’s likelihood of entering shelter, but neighborhood socioeconomic factors and building level characteristics, including building violations, residential units, rent regulation status, etc., are still not considered in determining vulnerability of entering shelter. The aim of this study was to identify the neighborhood and building level socioeconomic factors that indicate displacement and to utilize machine learning to effectively predict displacement and better target outreach. To do this, we used a random forest model to: 1) predict where tenants are most likely being displaced in New York City, 2) identify what building and neighborhood level factors are most predictive of tenant displacement, and 3) examine how these factors differ across space and housing type. Through this, we find that landlord consolidation is highly predictive of tenant displacement, and we argue that machine learning can be used as an effective tool for community groups to accurately identify and target areas of high displacement.