Social network analysis, predictive coding enlisted to fight fraud
Economic and budget realities have turned the spotlight on fraud, waste and abuse across federal, state and local government organizations, and agencies are employing new technologies that can detect collusive relationships and combat some of the more sophisticated fraud schemes.
New tools for combating fraud
Analytics/Predictive Modeling: Transactional fraud requires analytic technology to search other data repositories to identify validity of invoices and claims, services rendered and equipment delivery.
Data Fusion: Matching claims or billing data with information housed in other systems to identify suspicious activity.
Data Visualization: Visualization tools such as dashboards and modeling can help highlight instances and root causes of improper payments.
Social Network Analysis: Adds another avenue for detecting fraud rings and prior activities of suspected perpetrators.
Devices: Rising use of geo-spatial, mobile PCs, tablets, and smart devices with real-time data entry is being seen in programs and systems for tobacco tax, fleet management, and field inspection.
Source: Deltek, Technology Strategies for Federal Waste, Fraud, and Abuse
Technology that incorporates social network analysis, which helps establish connections and relationships between people, and predictive coding, which provides machine-learning techniques that help systems learn fraud patterns, are among the weapons agencies are enlisting to combat fraudulent payments, according to industry experts.
The Los Angeles County Department of Public Social Services is using social network and predictive analytics from SAS Analytics to identify potential fraud and prevent improper assistance payments. The Data Mining Solution for Child Care Welfare Fraud Detection, based on the SAS Fraud Framework for Government and SAS data mining techniques, debuted in May 2011.
Using the software, DPSS investigators detected two conspiring groups and mapped a network of participants and providers to display their relationships. Using the framework’s social relationship network capabilities, they created a display showing a web of complex relations linked by common telephone numbers and addresses. In one case, DPSS staff found a child care provider serving many participants working together in fraudulent activities.
Those types of results have spurred other state governments to adopt the SAS Fraud Framework for Government. Kentucky will use SAS Analytics with the state's new Health Benefits Exchange, analyzing eligibility and claims in Medicaid, food stamps and temporary assistance for signs of fraud. A statewide system will allow Kentucky to add new data sources and fight fraud in other areas in the future, according to Lori Flanery, Kentucky’s secretary of the Finance and Administration Cabinet and interim CIO. Michigan officials are deploying the SAS Fraud Framework in the state’s unemployment insurance and food stamp programs, with the goal of spotting fraud across all executive branch departments and programs.
The heart of the SAS Fraud Framework is an advanced analytics engine that monitors all of the transactions within any government program that is being monitored — Medicaid, unemployment, food stamps or taxation, said Greg Henderson, government practice lead for SAS Fraud and Financial Crimes Global Practice.The fraud framework can view all the information available about the entities involved in those transactions — providers, individuals, businesses, recipients or taxpayers — and use advanced analytical techniques to look for anomalies or high-risk payments.
A lot of information is buried in records that agencies may already have, Henderson said. The problem is that every government program has its own computer system, and fraudsters are able to exploit multiple programs because the systems are not connected. For instance, people enrolled in a child care subsidy program must be employed in order to be eligible for assistance, but someone collecting unemployment insurance technically is not working. When the two systems are linked, investigators often will find a handful of people collecting funds from both programs.
In another example, a medical provider might have a relationship with a durable medical equipment supplier (one who handles such things as prosthetic limbs, back or knee braces, oxygen concentrators, and wheelchairs) that involves a patient referral that is not appropriate or based on medical necessity. “We can establish those relationships and then study them using analytical techniques to look for relationships that are not normal,” Henderson said, explaining the role of social network analysis.
Earned Income Tax Credit, Federal IT Spending, Medicare, Medicaid, the Supplemental Nutritional Supplemental Program and Unemployment Insurance are some of the high-risk program areas where agencies are looking to prevent fraud and reduce improper payments and waste.
Improper payments in the federal sector have been slowly on the decline since President Obama signed an executive order nearly three years ago directing agencies to cut improper payments by $50 billion by the end of fiscal year 2012, according to the federal Payment Accuracy website. But more work needs to be done. The estimated loss from fraudulent payments and waste in 2012 was $19.2 billion for Medicaid and $2.7 billion for the SNAP, according to the website. And many reports indicate that fraud and identity theft are on the rise.
E-discovery tools, which have many uses, are increasingly being deployed to fight Medicare fraud, said Tom Kennedy, director of Symantec’s ClearWell eDiscovery Platform.
E-Discovery solutions let business users search and filter through large volumes of electronic documents for relevant information. Often used for general litigation support by attorneys, eDiscovery is now being applied by state governments to detect Medicare fraud as well as for environmental, public safety and securities use cases, Kennedy said.
The Clearwell platform sports numerous functions, including the cutting-edge predictive coding, which sometimes referred to as machine learning, he noted. Predictive coding involves training the computer to predict what documents are relevant and what is not. “You have to train it with a training set of data and then it goes out and predicts what is relevant to your case and what is not relevant,” Kennedy said.
Machine-learning is an emerging technology in the fight against online fraud. San Francisco-based Sift Science says it has developed an algorithm that uses machine-learning techniques to stay ahead of new fraud tactics as they are introduced into its customers’ networks.
Organizations have a growing list of technology weapons they can add to their arsenal to combat fraud including analytics/predictive modeling, data fusion, data visualization, social network analysis and geospatial-enabled devices, according to a recent Deltek report on federal fraud, waste and abuse.
Federal interest in analytics and big data solutions in particular will increase over the next five years and will aid in fighting fraud, waste and abuse. Big data technologies will help organizations glean intelligence from unstructured data, speed up analysis of massive amounts of data, and provide a method for accessing and matching data currently stored in disparate systems, according to Deltek analyst Angie Petty.
Rutrell Yasin is is a freelance technology writer for GCN.