sponsored

Why HPC Matters: Fraud Detection

Payment processors make credit and debit card transactions more foolproof with instant insights driven by high performance computing and machine learning algorithms.

1 whyhpcmatters frauddetection 1000
DELL EMC

Like the dusty streets of a Wild West town, the payment-processing industry is the setting for a never-ending shootout between good guys and bad guys. In this case, the good guys are armed with high performance computing systems that power machine learning algorithms designed to recognize and stop potentially fraudulent transactions initiated by the bad guys. This capability requires HPC systems that process enormous amounts of data on the fly with millisecond response times.

That’s the way it is at Mastercard. To identify and stop fraudulent transactions, the payment-processing powerhouse leverages machine learning algorithms running on HPC systems to process large data sets at lightning-fast speeds. Nick Curcuru, vice president of the big data practice at Mastercard International Inc., explains that the goal is to stop fraud in its tracks without disrupting or delaying legitimate transactions.

While that’s a challenging proposition for any company involved in retail sales, the scale at which Mastercard operates makes the problem mind-boggling. According to Curcuru, Mastercard has 2.2 billion cards in use in 330 countries. It processes 160 million transactions per hour, using machine learning algorithms and applying 1.9 million rules to examine each transaction. It all happens in a matter of milliseconds.

The horsepower for this fraud-prevention engine is a secure, Payment Card Industry (PCI)-certified Apache™ Hadoop® cluster based on HPC systems from Dell EMC. This fraud detection machine learning system uses supervised learning to look for established fraud patterns and unsupervised learning to identify emerging fraud patterns in real time.

With every transaction, the machine learning algorithms look at things like a cardholder’s buying habits, geographical location and travel patterns, along with real-time data on card usage — such as what they are trying to buy, where they are trying to buy it and what else they bought in the same day. Each transaction is analyzed in terms of the rules that relate to what a valid transaction looks like and what a fraudulent transaction looks like. The name of the game is to outsmart some really smart people with criminal intent.

“For us, it’s a question of how can we stay one step ahead, because the fraudsters themselves are just as smart as we are,” Curcuru explains. “They are graduating from some of the top universities worldwide. They are doing big data analytics themselves. This is a big business for fraudsters worldwide.”[1]

To keep ahead of the bad guys, the machine learning algorithms are always learning, so they can get continually better at what they do.

“It’s all about learning,” Curcuru says. “It’s not just one and done. The algorithms have to be constantly updated — in real time in some cases — so you’re constantly in a learning phase.” [2]

None of this would be possible without the enormous parallel processing power and fast throughput of today’s HPC clusters. These systems work together with data analytics tools and software frameworks, such as Hadoop, to enable the distributed storage, processing and analysis of huge amounts of data. And they often draw on the processing power of the GPU accelerators, as well as leading-edge many-core CPUs, such as those in the Intel® Xeon® and Intel® Xeon Phi™ processor families.

When everything is connected together with a high-speed, low-latency, next-generation fabric, such as the Intel® Omni-Path Architecture (Intel® OPA), you have a system that can complete millions of tasks in the blink of an eye — and that’s what it takes to detect and prevent the growing problem of payment-card fraud.

And just how bad is this problem? Really bad, according to The Nilson Report, a publication that covers the worldwide payments system. It predicts that worldwide losses to card fraud will exceed $31 billion in 2018. That’s an amount equal to 7.3 cents for every 100 dollars of total card volume. [3]

The good news: Today’s HPC systems are helping the payment-processing industry fight back in new ways. The combination of technologies that goes into fraud-prevention systems enables players in the industry to instantly analyze data while continually training algorithms to help them get better at recognizing fraudulent activities. 

The end result is a more trustworthy transaction experience for legitimate cardholders and more digital barriers to stop the criminals who try to exploit fraudulent cards. And that’s where the true value of HPC emerges for merchants and their customers.

For a closer look at how MasterCard uses machine learning to protect customer data, watch this video excerpt of Nick Curcuru’s interview with Dave Vellante at Dell EMC World 2016 in Austin, TX.

_____________________________________________________________

Making a difference with HPC

High performance computing touches virtually every aspect of our lives. HPC is making weather forecasts more accurate, cancer therapies more precise, fraud protection more foolproof and products more efficient. In this series of articles, we explore these and other use cases that capitalize on HPC and its convergence with data analytics to illustrate why HPC matters to all of us.

______________________________________________________________

[1] Interview with theCUBE conducted at Dell EMC World 2017.
[2] Interview with theCUBE conducted at Dell EMC World 2016.
[3]The Nilson Report, Card Fraud Worldwide,” October 2016.

Related: