Analysis

September 23, 2019

How machine learning is being used to tackle homelessness

When rough sleeping is flagged to authorities, the homeless person in need of help is rarely found by response teams. Can AI change this?


Kim Darrah

2 min read

Machine learning technology is transforming industries from healthcare to fashion. But can it also solve homelessness?

A team of data scientists backed by the Alan Turing Institute have been working with StreetLink, a homelessness charity, to look into how machine learning can help improve the decision-making process that goes on behind the scenes in homelessness support.  

Since 2012, StreetLink has run a nationwide system in the UK that provides members of the public with a way of notifying local authorities when they see a rough sleeper so that support teams can be alerted. 

Advertisement

But StreetLink has always faced the same problem: each alert consists of a jumble of details about a rough sleeper, such as their gender, clothes, location and condition. Often details are incomplete.

Rather than sending out teams to investigate each and every alert, the team at StreetLink must sift through the information and make a call about whether to send the alert to local response teams. 

The team assesses whether enough information was given to find the person and the level of urgency, based on the description given. Making both quick and good decisions is crucial so as to send support to where needed most. 

But as it stands, decisions are processed manually by a small team, and just one in seven alerts processed by StreetLink actually results in the homeless person being found. 

When there is a spike in the number of alerts - for example when a storm happens - the result is an overwhelming amount of work for Streetlink’s staff.

This is the kind of data problem where machine learning can help. 

By using information from past decisions, data scientists have created a machine learning model to automatically categorise alerts, giving Streetlink an immediate sense of which alerts should be prioritised. 

"If you can flag the top alerts automatically, then you can send those outreach teams out immediately,” says Austin Nguyen, one of the data scientists working on the project, which was arranged by the DSSG (Data Science for Social Good) programme in partnership with the Alan Turing Institute and Warwick University.

The model is yet to be implemented, but the hope is that it will help to free up resources so that volunteers at StreetLink aren’t under so much pressure to sift through large volumes of notifications. 

There are over 320,000 homeless people in the UK, according to research by Shelter, a housing charity, while rough sleeping in London is currently hitting record highs.

The model will be implemented for the first time in London over the coming months.