Predictive Risk Modeling to Identify Homeless Clients at Risk for Prioritizing Services using Routinely Collected Data

Abstract

AbstractFor most homelessness service providers, the number of clients who are eligible for long-term housing outstrips the availability. This study uses a cohort of housing assessments taken from a mid-size county in the US and machine learning methods to train a Predictive Risk Model (PRM) that identifies clients who would experience multiple adversities in the future. The PRM outperforms the Vulnerability Index-Service Prioritization Decision Assistance Tool (VI-SPDAT) in flagging clients at the greatest risk of adversities. The proposed method can be readily used by any Continuum of Care (CoC) that holds electronic housing assessments and service records.Keywords: Homelessnesspredictive risk modelhomelessness assessmenttriage toolprioritizing persons experiencing homelessnessVI-SPDAT Disclosure statementAuthors Rhema Vaithianathan and Chamari I. Kithulgoda report grants for their involvement in the project, developing predictive risk model to prioritize services for people experiencing homelessness in Allegheny County, PA, where they have taken research data for the present article.Data availability statementRestrictions apply to the availability of these data. Access to these data may be arranged on a case-by-case basis upon application to Allegheny County Department of Human Services.Ethics declarationThis research was reviewed and approved by the Auckland University of Technology Ethics Committee, reference number 20/176. The requirement for informed consent was waived by the committee.Notes1 This article refers to the VI-SPDAT surveys launched in 2015 due to its correspondence with the study data sample and the duration.2 https://www.cs.waikato.ac.nz/ml/weka/3 https://cran.r-project.org/src/contrib/Archive/glmnet/

Topics & Concepts