The accessibility of the Internet and the ability to commit fraud remotely creates an appetising prospect for fraudsters. The potential attack surface for miscreants is enormous; some 94 billion transactions were made for remote goods purchases in 2016, which is only a fraction of the total eCommerce landscape. Meanwhile, advanced security measures are increasingly being implemented to protect against fraud carried out at physical locations. It is for these reasons that fraudsters have developed, and are continually developing, new methods to illegally siphon cash over the Internet.
In response, service providers are using machine learning to perform risk assessment based on a user’s behaviour, either on a web site or using an app. The aim of such techniques is to reduce the number of instances where rules-based systems would either reject a transaction, or hold the transaction pending a costly manual review.
The benefits of using machine learning models to determine a transaction in terms of its risk are numerous:
- An ability to leverage a complex relationship between multiple data inputs rather than a rigid rule structure;
- The model is able to evolve over time and improve as more data is input and patterns are understood;
- New trends in fraudster approaches can be added to the model to react to prevailing market conditions;
- Models are able to use behavioural inputs, such as mouse movement, touch screen behaviour and various other indicators to separate genuine user behaviour from fraudulent behaviour.
- Algorithms are able to detect patterns in data that would normally be hidden, thus offering the potential to save human-hours.
Our latest research into the Online Payment Fraud market, Online Payment Fraud: Emerging Threats, Key Vertical Strategies & Market Forecasts 2017-2022
, explores rising challenges in defending against fraud, and the opportunities open to service providers.