This is a guest post.
As e-commerce has grown more widespread and sophisticated in recent years, so too has the attendant risk of fraud. In the past five years, roughly one in three Americans have had their credit card information stolen, and fraud attacks continue to grow more complex and difficult to counter.
However, as technology makes fraud more pervasive, it also provides more powerful tools for consumers to stay protected. Machine learning is one such tool, and it promises to radically alter the balance of power between individuals and the criminals who seek to defraud them.
The Growing Threat of Modern Fraud
Though the costs associated with payment fraud are well-known, it only represents one of many dangers posed by modern digital fraud. Identity theft, account takeovers and data breaches have become increasingly common, and in addition to the financial costs, consumers are also subject to other damages. After a series of large data breaches at retailers like Target and Home Depot, and more recently health insurance carrier Anthem, 45 percent of online households have reported that they’ve stopped engaging in purchases, financial transactions or other activities over privacy and security concerns. Fraud attempts also pose the risk of legal issues, loss of sensitive data and other tangential costs that can cause lasting harm.
Tom Galvin, Executive Director of the Digital Citizens Alliance, spoke on the issue saying, “Americans want their leaders – whether they work in government or at the tech companies that bring us technology – to step up and combat this epidemic of online crime and risk. Americans deserve to feel safe whether they are shopping in a mall or on Amazon. They shouldn’t have to worry about someone stealing their credit card information, whether it’s at a restaurant or on Etsy.”
Digital Identity Tracking
One of the greatest challenges faced by online companies is verifying the identities of different shoppers and the authenticity of their transactions. Machine learning offers new tools, including a Segment-of-One approach that facilitates quick and detailed tracking of individual consumers across a company’s network. This allows businesses to gain better insights into a person’s behaviors and readily identify and flag potentially suspicious accounts and transactions. These systems are capable of processing tremendous volumes of data far more quickly and effectively than humans, refining and improving results as more and better datasets become available.
Fighting Fraud with Advanced Analytics
Online retailers spend about seven percent of their total revenue on combatting fraud, using both detection and prevention strategies in an attempt to stay ahead of innovative fraudsters. Among the many tools afforded by machine learning technology, three broad analytical methods have emerged: descriptive analytics, predictive analytics and social network analysis.
- Descriptive Analytics
Also called unsupervised learning, descriptive analytics can review datasets to identify transactions or behaviors that fall outside of “normal” – typically defined by average consumer behaviors over a specified period of time. This is done through a combination of techniques including association rules, peer group analysis and clustering.
- Predictive Analytics
Predictive analytics, or supervised learning, uses historical datasets containing known instances of fraud to build sophisticated models that can then be applied to real-time detection. Because it relies on known fraud cases, predictive analytics cannot detect previously unknown fraudulent activities, but it leverages the power of logistic and linear regression, neural networks and random forests to analyze large datasets and create highly sophisticated detection models.
- Social Network Analysis
One of the most popular fraud detection tools is Social Network Analysis (SNA), which employs community detection, featurization and other visual and analytical tools to detect patterns and connections across broad networks. This technique helps to gain insights into not only individual actions and actors but also the connections between them, providing an entirely new look into how fraud occurs and how it can be detected and prevented.
Smarter Digital Payment Solutions
In addition to combatting payment fraud in the online commerce space, machine learning and other artificial intelligence technologies are creating the opportunity for new digital banking and payment solutions. Currently about 73 percent of financial trading is executed by sophisticated AI algorithms; so-called “robo-advisors” have taken on a large role in portfolio and wealth management and there’s a growing consensus that loan underwriting will soon also be done largely – if not entirely – by advanced algorithms.
Additionally, the Internet of Things is expanding to encompass an “Internet of Payments.” Voice-interface home hubs like Amazon’s Echo and Google’s Home make it possible for consumers to conduct banking transactions and via digital assistant. Certain wearable products, like smartwatches and fitness bands, also offer payment services. According to a Paga survey, 24% of respondents said they believe consumers will commonly make transactions using appliances or smart home controllers in the next 2 years. This will pick up as the machine-learning elements of voice technology improve, and as biometric security integration becomes more widely prevalent.
Machine Learning in Action
When it comes to applying the latest machine learning technologies, several companies are leading the charge. Mastercard’s Decision Intelligence platform uses machine learning techniques to enhance the accuracy of transaction approvals, weeding out fraudulent activity and reducing the number of false card declines. Memorial Health System, one of the largest public healthcare networks in America, uses IBM’s AI-powered Big Data analytics platform in its VETTED system to evaluate vendor activities, spot fraudulent behaviors and improve service delivery. PayPal is also ahead of the pack, using artificial neural networks and other tools to keep its fraud rate exceptionally low.
The nature of online fraud will surely continue to evolve to keep pace with the ever-accelerating development of technology, and everyone must stay ahead of the curve in order to protect themselves. The ramifications of failing in this regard can be gravely serious, making it essential to understand and embrace the power of machine learning and other artificial intelligence technologies to detect and prevent fraud in all its forms.
Beth Kotz is a freelance finance and tech writer with a strong interest in artificial intelligence and its vast potential for society. A graduate of DePaul University, she is currently based out of Chicago, Ill. You can visit her website here.