Preventing Eviction-Caused Homelessness through ML-Informed Distribution of Rental Assistance

TL;DR

This study demonstrates that machine learning can transform homelessness-prevention policy by shifting rental-assistance allocation from a reactive, first-come approach to a proactive, risk-based system. Using 2012–2023 county/state administrative data, the authors train multiple models to predict homelessness-service interactions within $12$ months for eviction-facing tenants, prioritizing the top $100$ predicted cases. The best models (RF and LR) outperform heuristic baselines by roughly $20\%$ in precision@$100$ and $10\times$ compared to the current process, while identifying $28\%$ of individuals who would otherwise be overlooked and become homeless. Field-validation plans include Shadow Mode Deployment and an eventual RCT, with careful attention to equity (race and gender), data leakage, and policy communication to ensure real-world impact. The work provides practical lessons on data needs, model design, evaluation, and field validation for evidence-based decision-support tools in housing stability and beyond.

Abstract

Rental assistance programs provide individuals with financial assistance to prevent housing instabilities caused by evictions and avert homelessness. Since these programs operate under resource constraints, they must decide who to prioritize. Typically, funding is distributed by a reactive or first-come-first serve allocation process that does not systematically consider risk of future homelessness. We partnered with Allegheny County, PA to explore a proactive allocation approach that prioritizes individuals facing eviction based on their risk of future homelessness. Our ML system that uses state and county administrative data to accurately identify individuals in need of support outperforms simpler prioritization approaches by at least 20% while being fair and equitable across race and gender. Furthermore, our approach would identify 28% of individuals who are overlooked by the current process and end up homeless. Beyond improvements to the rental assistance program in Allegheny County, this study can inform the development of evidence-based decision support tools in similar contexts, including lessons about data needs, model design, evaluation, and field validation.

Figures (9)

• Figure 1: Inclusion/exclusion criteria for homelessness and eviction. Indicators for homelessness in ACDHS data include clients' interactions with homelessness services, such as staying in a shelter. We also include clients enrolled in rehousing programs that have not moved in as of the prediction date, as being homeless is a prerequisite for enrollment.
• Figure 2: LR and RF outperform all baselines: Precision@100 over time shows that out of all baselines, B1: Previously Homeless performs best, but LR and RF perform better for all splits outside of the moratorium.
• Figure 3: Percentage found of missed group, the individuals who become homeless having not contacted Allegheny County Link or received rental assistance.
• Figure 4: False positives are still vulnerable: among these tenants, we see that homeless service utilization and mental or behavioral health crises are common beyond 1 year.
• Figure 5: Temporal validation