Fighting Fraud with
Machine Learning

Organisations are continually searching for better ways to tackle fraud without implementing cumbersome security measures that alienate customers. With an ever-evolving landscape of new and faster payment methods, companies need to recognise transaction anomalies quickly to facilitate real-time decisions.

Fraud detection tools that leverage machine learning can enable companies to dramatically improve accuracy and efficiency. In the report Digital Payment Fraud in High Growth Markets by Forrester Consulting, commissioned by LexisNexis® Risk Solutions, customers ranked artificial intelligence (AI) and machine learning usage as the most important components in a successful fraud management solution.

Learn more from Pratik Choudhary, Business Development Manager and Sheldon West, Head of Data Science and Analytics for Fraud and Identity in EMEA Professional Services, as they delve into the workings of machine learning and explore its significant impact in combating fraud.

How do companies use machine learning to address payment fraud?

Customers generally employ two main practices of machine learning, supervised and unsupervised. In supervised machine learning, a model is built using data inputs that are paired against a label detailing whether the payment was genuine or fraudulent and ideally including the type of fraud such as scam versus account takeover.

When data scientists create a model with the right data, labels, and features, the model learns patterns. With that information, the model can sift through more data than a human and sort safe payments from potentially fraudulent ones. Companies benefit from actionable insights in real time with fewer false positives.

In unsupervised learning, the model does not require a label for every single event or payment. Instead, it leverages algorithms to find segments of interest within the data. This approach is particularly helpful in anomaly detection since every observation is not forced into a cluster. It is especially useful for identifying outliers against a user’s baseline profile, as seen in LexisNexis® BehavioSec® User Profiling, which recognises legitimate users by analysing their unique behavioural patterns. This allows the system to identify those data points that differ significantly from their peers, warranting further review. Unsupervised learning is particularly effective in cases where fraud may not be obvious.

How do machine learning models optimise customer experience, trust and risk? 

This technology can be applied in many ways. One common application is to help companies and banks distinguish between the type and level of risk. For example, when a consumer makes a large purchase, the model may prompt a step-up to multifactor authentication. While this approach is effective at mitigating traditional fraud types such as account takeover, the same safeguards don’t work for every situation.

Another useful application of machine learning is recognising risks from social engineering scams where signals are more subtle and therefore more difficult to detect. Machine learning’s high degree of flexibility makes it particularly effective in these cases.

For example, in the case of a bad actor trying to compel an individual to transfer his own savings to an unauthorised account. Since it is the legitimate user using his own device from his own location, a step-up won’t work making this type of fraud particularly difficult to identify and manage.

However, we know that this type of situation requires an open channel of communication between the fraudster and victim – perhaps a phone call. Machine learning allows us to blend innovative features like active call detection with other signals that lead to precise alerts demanding a more nuanced intervention strategy.

How does machine learning apply to ecommerce fraud prevention strategies?

Within the world of ecommerce, fraudsters tend to be more opportunistic targeting refund policies, return policies and other commercial gestures that companies offer to attract consumers. In these instances, machine learning can help quantify the “trust to risk spectrum.”

For instance, it can identify most well-intentioned users, allowing the business to provide a seamless instant rebate while alerting the company to detected irregularities. This capability to hinder bad actors while ensuring a smooth transaction for legitimate customers can be seen as the 'golden egg' of enhancing customer experience and fostering loyalty.

How can businesses leverage the power of machine learning to fight fraud?

Machine learning is a powerful tool based on statistics and math, but it does require a “few key ingredients” to optimise effectiveness. First and foremost is reliable and broad-based data. At the same time, a company’s data scientists should have a solid understanding of fraud and financial crime so they can build appropriate models. And while technical expertise is important, the ability to think creatively is an equally critical attribute. Lastly, collaboration such as crowdsourcing of intelligence among industry participants can provide beneficial insights about existing and emerging risks. In short, fraud keeps evolving and continuous innovation is the key to keeping pace.

Customer experience is the focus, so the overriding goal is always “to ensure a great experience for those who deserve it.”  The power of machine learning lies in its customisability. LexisNexis® Risk Solutions builds tailored solutions that help companies separate bad actors, scams and fraud events from legitimate interactions with end-consumers.

Click here to listen to the Identity Trust Pulse podcast with Pratik Choudhary and Sheldon West.