Evaluating AI-Enabled Eviction Prevention for Low-Income Tenants

The pilot will develop and evaluate AI-enabled community outreach for preventing low-income tenant eviction. Housing insecurity and homelessness threaten the health and social inclusion of marginalized populations around the US and the world. Research consistently demonstrates the utility of homelessness prevention; one well-documented local initiative offers landlord-tenant mediation at court before an eviction filing. Yet, diversion programs struggle to scale and often fail to provide timely and tailored support that prevents housing crises from escalating. The present study builds upon an ongoing research partnership with local housing advocates to leverage AI with administrative records to prevent evictions. Machine learning modeling will be enhanced through newly available multimodal data that captures characteristics of tenants (e.g., eviction histories), properties (e.g., building age, code violations), property owners (e.g., property count), and neighborhoods (e.g., calls for social services) over time. Models will be trained with historical data to predict the likelihood that a tenant will face eviction in court within the next three months. In addition to cross-validation on a held-out test set, online surveys with a stratified random sample of 1,111 households identified as at-risk will assess the predictive accuracy of formal (i.e., in court) and informal (i.e., extra-jurisdictional) evictions. Surveys will also capture tenant perceived support needs and existing service connections, while insights will be presented back to housing advocates to inform the design of coordinated eviction diversion. Previously built simulations will evaluate the efficiency and equity of AI-enabled community outreach, as well as potential unintended consequences for marginalized tenants.