Using Machine Learning to Connect Rough Sleepers to Local Services

18.03. | 16:15 - 16:45 | data.stage

Homelessness and rough sleeping comprise a global issue that subjects a population to a host of societal and health-related pressures spanning abuse, illness, poverty and alienation. How can machine learning help the homeless? Homeless Link is a UK membership charity organisation that operates a platform called StreetLink, allowing members of the public to submit an alert when they see someone they suspect is sleeping on the streets, or rough sleeping. The alerts are subsequently passed on to a team of volunteers who manually review each alert for quality and descriptive power before a referral is dispatched to relevant, local service providers. This process takes a significant amount of time, which can have detrimental effects on the level of service or even be impossible to deal with during surge times in extreme weather and the winter months. We partnered with Homeless Link to develop a machine learning model to assess the quality of alerts. We generated a prioritised list of alerts with the potential for the top alerts to be automatically sent to StreetLink as referrals. Our simulation suggests that with our model the rate at which rough sleepers are found after referrals would increase by 18%. Our work also sheds light on the features that are predictive of the rough sleepers‘ likelihood of being found. This work demonstrates the power of a data-driven approach in furthering a charity organization’s mission of reducing the number of rough sleeper.


Harrison Wilde (University of Warwick / Alan Turing Institute)
Austin Nguyen (TripAdvisor)
Lushi Chen (The University of Edinburgh)