Modeling and predicting individual transitions within the homelessness system

This study focuses on how individuals navigate the homelessness system over time, with the ultimate goal of securing stable housing. Administrative data collected by homeless service providers are used to infer the unobserved underlying network of homeless services. A similarity score between the ordered sequences of services that individuals receive is proposed. The score leverages the structure of the inferred network in addition to historical observations to identify individuals with similar trajectories. In doing so, the service an individual will be assigned to next can be predicted. Extensive experiments show that the proposed approach performs well not only on predicting exit from the system, or simply guessing high frequency services (as most baselines), but is also successful in less frequent scenarios. Building a model that learns to replicate the dynamics of the existing system is the first step toward developing computational methods to maximize outcomes (i.e., ensuring that as many homeless individuals as possible secure stable housing).