"The challenge is significant. The NAO said the DWP's IT systems are not fully integrated, preventing staff from accessing complete claimant information. The department is working to develop an application to provide a single view and told the auditor that scaling up would need cross-government data standards to enable inter-departmental data sharing. Denmark, which has introduced interoperable IT systems and government-wide data standards, has around 100 anti-fraud machine learning models."
"The NAO found fairness issues. Applicants aged 45-plus and non-UK nationals were more likely to be flagged but less likely to have claims refused. The DWP assessed this for only one of nine protected characteristics under equality law - age - as it lacked sufficient data on the others. Despite this, the model is three times as effective than random sampling and will remain in use while being improved."
"The UK government's Department for Work and Pensions (DWP) has saved £4.4 million over three years by using machine learning to tackle fraud, according to the National Audit Office (NAO). DWP's current machine learning work focuses on Universal Credit, which is replacing a number of legacy benefits. Since May 2022, a model has flagged potentially fraudulent hardship payment advance claims for human review rather than automatic rejection."
The Department for Work and Pensions saved £4.4 million over three years by using machine learning to detect fraud. The department's machine learning work currently focuses on Universal Credit and includes a model that flags potentially fraudulent hardship payment advance claims for human review rather than automatic rejection. Expansion is limited by fragmented IT systems and the absence of cross-government data standards, prompting work on a single-view application. Fairness issues emerged: applicants aged 45-plus and non-UK nationals were more likely to be flagged but less likely to have claims refused. Four further Universal Credit models are in development targeting undeclared self-employment income, financial assets and undisclosed partners.
Read at Theregister
Unable to calculate read time
Collection
[
|
...
]