The Digital Art of Detecting Fraud

Print Friendly, PDF & Email

In this special guest feature, Christopher Hillman, Principal Data Scientist in the International Advanced Analytics team at Teradata, discusses the significant progress being made in the area of fraud detection through use of machine learning by providing new insights into existing problems. Chris has worked on a large number of Data Science projects across the Teradata International region focusing on Big Data technologies. He has spoken regularly at leading conferences including Strata and Hadoop. One of his biggest achievements has been working with two other collaborators to establish the Art of Analytics practice, promoting the value of producing striking visualizations that draw people in to the work we are doing, while retaining a solid business-led foundation. As well as working for Teradata, Chris has been studying for a PhD part-time at the University of Dundee since 2012, expecting to graduate in 2018. His research concerns applying Big Data analytics to the data produced from experimentation into the Human Proteome.

Fraud is a game of cat and mouse — but it’s one that businesses can stay ahead in if they choose intelligent solutions to this pervasive problem.

To date, it’s been difficult for fraud teams to tackle the tidal wave of scams aimed at exploiting their company, with criminals constantly evolving their tactics to outpace enterprises. The Coalition Against Insurance Fraud pegs its industry’s annual total of claims at $80 billion — about half a percent of the United State’s gross domestic product. Insurance fraud has become so widespread that it accounts for 5 to 10 percent of claims costs in both the United States and Canada, with some insurers pegging it closer to 20 percent.

The task of identifying these cases cannot be relegated by humans alone with the kind of scope and scale necessary. To stay ahead of the overwhelming number of malicious claims in the insurance industry, companies must turn to a technology that can handle massive amounts of data, detecting patterns that provide clues to when fraud is occurring. And for this technology to be really effective, it needs to be more than spreadsheets of facts and figures.

Data visualization serves as a digital form of art that can intuitively display for insurance fraud teams how all their claims are connected, which bad actors they should zero in on and which claims are likely authentic. By visualizing the connections detected between claims using data such as phone numbers, addresses and other typical insurance claim data, analytics can detect traits of fraud, like multiple cases that include the same information, or a claim made just after policy limits are increased, for example. The larger a cluster or graph appears, the bigger the fraud ring. This kind of visual analysis allows companies to catch claims earlier in the process, even before a fraud investigation team looks at it. Then, agents can focus on highly suspect cases, versus ones that are likely legitimate.

The backbone of this type of data visualization technology is machine learning. Machine learning allows computers to create models based on the information they are fed in real time, “learning” as it goes without needed to be specifically programmed to do so. As the system is exposed to new data, it will get better at detecting these patterns based on the features that hold more valuable information. It is intelligent enough to be able to parse through text from call center agents that handle claims, analyzing the language for indications of fraud.

The data visualization platform has the ability to utilize a series of analytic techniques in conjunction to gain insight into fraudulent cases, like text mining, graph analysis, cluster analytics and predictive analytics. This approach creates a more flexible fraud detection algorithm and additionally agile ways to visualize the data.

Data visualization combined with machine learning can help employees better understand their fraud protection business processes in a new way. Aside from providing new insights into existing problems, it’s also a new way to drive business value, avoiding costs and liability. Even a small slice of the $80 billion pie amounts to a huge recovery in assets.

One in three insurers don’t feel adequately protected against fraud. By leveraging visualizations of the output from machine learning algorithms, enterprises can dig their fraud investigation teams out of their endless backlog of cases. Knowing which claims are likely illegal before they even start reviewing with the help of intuitive data visualization will help insurance companies stay ahead of this evolving threat.


Sign up for the free insideBIGDATA newsletter.



Speak Your Mind