Smart Homelessness Service Provision with Machine Learning
Charalampos Chelmis, Wenting Qi, Wonhyung Lee, Stephanie Duncan
Abstract
Homelessness presents a long-standing social problem for nearly every community across the world. A key goal of homelessness service provision is to reduce the number of individuals who experience repeated episodes of homelessness. The goal of this work is to determine the feasibility of an automated recommendation system designed to carefully match individuals to homelessness service facilities when they first experience homelessness. Specifically, machine learning methods are used to recommend the exact service facility that a homeless individual can benefit from among other numerous homeless-serving organizations in the Capital Region of New York, based on individual time-variant (e.g., monthly income, age) and time-invariant (e.g., race, sex) features. The data used in the study span a total of 38, 800 individuals seeking federally funded homelessness assistance from 2005 through 2019. The best performing method achieves an accuracy of 81.5%.