How financial institutions can improve their current approaches to preventing fraud

The battle to prevent fraud in real time is an ongoing process. Chris Dorrington from Featurespace™ suggests ways in which financial institutions (FIs) can improve their current approaches. 

Financial institutions today are able to adjust risk scores as their customers’ behaviours change. This has been made possible by the use of risk-driven modelling, which adapts to shifts in spending. These models can help detect fraud.  

Deep learning has accelerated the development of this technology. The following is an introduction to deep learning and how it is used to prevent fraud.  

Machine learning  

Many banks’ fraud prevention tools today are built with some form of machine learning (ML). Machine learning has done an excellent job at improving how banks detect and prevent fraud. It has also given financial institutions the tools they need to not only protect their customers and save money, but also to disrupt fraudsters.  

Machine learning-powered software can work in real time to predict the fraud risk of a given transaction. A traditional machine learning model works by applying logic to raw data to identify fraudulent payments across a large volume of transactions. The key word here is 'logic.', but whose logic?

 Most banks employ teams to investigate and react to fraud scenarios by introducing algorithms into the machine learning model that can improve decision making. However, as we all know, only applying human logic can mean that our own biases inadvertently creep into the model's decision making. In turn, as soon as a traditional model encounters a novel problem, its performance is hindered because the model is restricted by the information its creator provided. Ultimately, that’s what holds traditional machine learning approaches back.  

Deep behavioural learning   

Deep learning, a subset of machine learning, helps to solve this problem. This is done by giving models the ability to extract information from raw data and make decisions independently of any pre-programmed decisions from an analytics team. In transaction monitoring, the models are empowered to identify signals from the past as well as the present, effectively creating models with memory.  

Deep learning technology in fraud detection works similarly to natural language processing, a technology that, with sufficient context, can be used to understand sentences and predict upcoming words. With our Automated Deep Behavioural Networks, it is now possible to understand what an individual customer’s spending should look like, and identify the spending behaviours that are anomalous.  

Fraudsters have reacted and begun to change their tactics. When you close one door, another opens.   

Fraudsters’ new approaches centre around customers as the weak point. Individuals are the target of sophisticated social engineering activities. UK Finance’s latest fraud figures, published last month, show that the value of all APP cases in 2021 was up 39 per cent compared to 2020, at £583.2 million. These losses now surpass the amount stolen through card fraud in 2021 (£524.5 million).

Banks have always had the difficult task of balancing fraud prevention and customer service. Deep learning adds new urgency: It is now more important than ever for banks to understand the benefits of automation to predict their customers’ behaviours and mitigate fraudulent transactions. 

To discover more insights on using machine learning to detect and prevent fraud in real time, download our whitepaper 'Machine Learning for Fraud Detection: What Risk Professionals Need to Know' here.



Area of expertise: