Federal government agencies are faced with an immense amount of data, with more pouring in every second. With so much information, keeping track of it all can be extremely challenging, particularly when there are bad actors seeking to take advantage of the data overflow.
Currently, federal agencies maintain more than 2,200 federal assistance programs that distribute more than $2 trillion per year in subsidies. Of that, approximately $125 billion is lost to fraudulent or improper payments each year.
So, how can agencies confront the issue? Artificial intelligence and machine learning may be the keys.
To discuss using AI and machine learning to better manage data and confront fraud, GovLoop spoke with Randall Knol, IT Specialist, Commerce Department Census Bureau Demographic Surveys Division Application Support & Innovation Branch, Dave Vennergrund, Data and Analytics Service Area Director, General Dynamics Information Technology, and Karen Painter, Sr. Program Manager, General Dynamics Information Technology, during the virtual summit, “Gov Tech Trends to Pay Attention to in 2019.”
Knol began by describing the challenges faced at his agency.
“The big project right now is creating a data lake because census is the repository of the federal administrative data,” Knol said. “Unfortunately, that data comes in a wide variety of formats and isn’t easy to use. We can do it manually, but it’s slow and not very effective, so we’re looking for machine learning to help with our data linkages and address the challenges that we have with time and money. We want to be able to analyze data quality more quickly and easily. Quality is the most important thing here.”
With this challenge, there are many opportunities for bad actors to attempt to take advantage of the system – particularly in the case of fraud. Vennergrund provided a number of examples of fraud cases before diving into how they can be confronted through better data management with AI and machine learning.
General Dynamics Information Technology collects information and uses predictive models to find potentially fraudulent behavior and reduce fraud, waste and abuse.
“With all of the data that we were able to receive across a number of programs, all these data sources provide a rich view that we can use to build predictive models,” Vennergrund said. “[Electronic Fraud Detection System] has prevented $10 billion in fraudulent tax refund claims.”
He added that the system uses AI and machine learning to conduct statistical analysis and crack down on fraud.
Painter provided the specific example of Medicare and Medicaid, which are regularly targeted due to the sheer amount of data produced.
“[In healthcare,] billions of transactions can make detection very difficult,” Painter said. “There is a lot of data to sift through. In addition, healthcare isn’t black and white; each patient has specific needs. The Department of Justice initiated the fraud strike force in 2007 and has well over 2000 indictments today. Our team pulls in data from many data sources to look for anomalous trends.”
But fraud schemes aren’t stagnant. With constantly changing threats, agencies have to stay one step ahead, which is where AI and machine learning predictive analytics can help.
“Fraud schemes are continuously changing, so we focus on identifying new schemes,” Painter said. “We’re continually evaluating the performance of our models.”
Vennergrund provided a few best practices for fraud prevention to end the session.
“Use all of your data. Leverage cloud and analytical services. Apply deep learning methods. And use AI-assisted tools to discover data assets and relationships.”
If you missed this virtual summit, make sure to sign up at this link to be notified when future summits are happening!