AUT's Centre for Social Data Analytics (CSDA) has deployed a world-first machine learning tool for a county in the United States, which accurately prioritises requests for housing help.
The Allegheny Housing Assessment (AHA) tool has been delivered as part of CSDA’s ongoing data science for social good collaboration with the Allegheny County Department of Human Services, Pennsylvania.
The AHA tool is used by county staff when they answer calls from people at risk of becoming homeless. Because the county has a limited supply of long-term supportive housing, requests for help must be prioritised, and housing offered only to those who need it most.
The AHA replaces an actuarial tool that has been used in Allegheny County for many years that, while widely adopted in the US, is not locally validated and relies on self-reported personal information from the client.
“The idea of supportive housing is to reduce the harms that come with homelessness, so we have built the AHA tool to prioritise those who are greatest risk of those harms - mental illness, jail time and health emergencies," says lead data scientist Dr Chamari Kithulgoda.
“When we help the county to direct resources towards those at greatest risk, we boost the potential positive impact of supportive housing across the board.
Professor Rhema Vaithianathan leads the AHA research team at CSDA.
“The AHA offers the superior accuracy of a machine learning tool trained on local data but also the fairness of a tool that has been built with strong ethical oversight,” she says
The county commissioned an independent ethical review of the AHA that informed its development and deployment. Ethicists found that the tool was accurate and did not introduce any disadvantage for vulnerable groups, but also made recommendations on key aspects including regular remodelling, fairness analysis, explainability and worker training that were welcomed by the county in its response to the ethical review.
Allegheny County intends to commission an independent evaluation of the AHA in 2021, which will provide evidence of its impact on the quality of supportive housing decisions in Allegheny County.