Skip to content

Rise Of The Machine (Learning) and Fraud Prevention

Machine Learning and Fraud Prevention

Machine learning, as a sub-discipline within computer science, is primarily concerned with the discovery of patterns in data through algorithms that can learn from and make predictions on that data. These algorithms operate by building a model based on example inputs, which can then make data-driven predictions or decisions. So what, exactly, does this have to do with beating fraud in the real world?


From big data to actionable insights

Machine learning algorithms can be used to build predictive models, which assess the probability of a transaction being fraudulent. This is achieved by combining historical transaction data from fraudulent activity, together with information from genuine customer transactions, delivering very tangible results in fraud detection.

The ability of machine learning algorithms to extract meaning from complicated data means that they can be used to identify patterns and highlight trends that are simply too complex – or require too much data to be crunched - to be noticed either by human fraud analysts or through other automated techniques. By running specific algorithms, and using them to make automated decisions or generate alerts for suspicious activity, it is possible to save manual review time, reduce the number of false positives, and quickly stop fraud attempts.


Even models need rules

Models learn by example – so the more valuable, complete, and relevant data they are fed and trained on, the more accurate they will be. Access to merchant, payment provider and FI data across sector and geographies helps to enhance and train machine learning models. From the perspective of retail fraud, there are two primary models that need to be considered.

Sector Models:

Sector models are developed based on data from multiple merchants within a given sector. Fraudsters operate across retailers and geographies, so a global view helps to ensure higher predictive performance.

Tactical Models:

Tactical models can be developed for a specific segment, data strata, or merchant. The customized nature of the modelling process means that implementing tactical models comes with a higher cost, but they can enable merchants to achieve higher fraud detection rates at lower false positive rates.

Combining either of these machine learning models with a sophisticated rules engine allows fraud strategies to be fine-tuned to the needs of individual merchants. A rules engine needs to be flexible, so that merchants can quickly adapt to emerging fraud trends, as well as respond to business changes such as entry into new markets or channels. Ideally, this is done in such a way that it is not necessary to ‘retrain’ the machine learning model.


Who’s in the driver’s seat?

The successful use of machine learning within the payments and fraud space requires dedicated data scientists, who can develop the models, and also apply the use of algorithms in detecting fraud. Data scientists, however, cannot develop these capabilities in isolation, and work closely alongside risk analysts and software engineering teams. Most importantly, such cross-functional teams must strive to constantly enhance the relevance and performance of machine learning models, to ensure they remain effective as the payments landscape changes. And the ability to navigate this evolving landscape – despite the increasing power of machine learning – still requires a distinctly human touch.

 

ACI Worldwide will participate in a panel discussion on Machine Learning at MRC London, April 25th at 4:15pm, as well as showcasing online fraud management and payments capabilities.

 

Panel Discussion Machine Learning