arrow left facebook twitter linkedin medium menu play circle
November 24, 2020 - Claire Zhou

Unsupervised Machine Learning to Reduce False Positives in Fraud Detection

False positives in fraud detection are a growing problem that can cost revenue and customers. Incorrectly declining a credit card creates unnecessary friction during the buying process, but traditional fraud detection methods often make this costly mistake.

There is an answer to this growing problem: unsupervised machine learning (UML). 

UML uses artificial intelligence to decrease costly false positives. Here’s how it can help your business.

Fraud + False Positives = Big Revenue Losses

Imagine you’re a customer making a routine credit card purchase when your card is declined. The problem isn’t that you’re late on payments or your card maximum was exceeded. Instead, a computer algorithm incorrectly identified you as a fraudster. This happens to around 33 million Americans every year. 

It’s a frustrating and embarrassing event for a loyal customer, and it costs companies $118 billion in blocked sales every year. In addition to the lost opportunity cost, many people end up never using the card again or switching carriers—anything to avoid the stigma of a false decline. 

It’s an embarrassing scenario that’s happening more often as banks are forced to become more aggressive in their efforts to combat fraud. Four in ten Americans have had a transaction blocked or questioned by their card company, according to creditcards.com

What happens when a company gets it wrong and flags a legitimate transaction as fraudulent? The studies show that even one false positive can lead to customer loss:

  • 39% of customers stop using their cards after even one false positive.
  • Another 25% put the card in the back of their wallet and use it less.

What’s even worse is that you stand to lose, not just the customer who was incorrectly labeled a fraudster, but other customers as well. How? The simple answer is that news of a bad customer experience spreads like wildfire in today’s always connected world.

Consider these statistics:

It’s an untenable situation that costs you big money each year. 

Traditional Fraud Detection Versus Unsupervised Machine Learning Fraud Detection

Like all technologies, fraud detection software has undergone an evolution. The first evolution was traditional detection that relied on outdated rules engines that check fraud activity against a rules list and combine the rule with logic to stop or allow the activity. These manual processes simply cannot keep up with fraud incidents powered by artificial intelligence (AI) algorithms. 

The next evolution in fraud detection applies supervised machine learning to fraud detection. This process leverages AI algorithms to learn from the existing data and apply that knowledge to new activity. Supervised machine learning is the most common fraud detection used today but, used alone, it carries the risk of high false positives that can ultimately hurt your bottom line. There are a number of other drawbacks to these tools that make them a slower and more unwieldy approach to the fast-evolving, sophisticated fraud techniques of today. Supervised machine learning relies on legacy knowledge, not new or unknown fraud techniques.

Fortunately, today there is another iteration of fraud management that applies unsupervised machine learning to detect fraudsters. Unsupervised machine learning (UML) is suitable to combat the sophistication of today’s cyber criminals. It works to discern patterns in large unstructured data sets and is the most proactive and efficient approach to detecting fraud before it can happen. UML helps companies avoid the damaging false positives that cost customers. 

DataVisor has successfully applied UML for our clients to lessen the impact of:

  • Application fraud
  • Bot attacks
  • Money laundering
  • Promotion abuse
  • Transaction fraud

UML works to prevent fraud while lessening the impact of false positives. DataVisor’s UML-driven software can save your company by preventing fraud and by stopping false positives that cost you revenue. To learn more about UML and its place in a comprehensive fraud protection strategy, check out our infographic entitled “5 Common Misconceptions about Unsupervised Machine Learning in Fraud Detection.” 

Photo of Claire Zhou
about Claire Zhou
Claire is a Product Marketing Manager at DataVisor with over 5 years of marketing experience in security and fin-tech. She is passionate about empowering enterprise customers with AI-based solutions. Her expertise spans data analytics, cybersecurity, and fraud prevention. Claire has an MBA from UCLA.
Photo of Claire Zhou
about Claire Zhou
Claire is a Product Marketing Manager at DataVisor with over 5 years of marketing experience in security and fin-tech. She is passionate about empowering enterprise customers with AI-based solutions. Her expertise spans data analytics, cybersecurity, and fraud prevention. Claire has an MBA from UCLA.