How data science has changed fraud detection
We live in a data driven world where many of the actions we take, as a person or a business, can be recorded and quantified. Until recently, businesses did not have the computing power or the expertise to take advantage of this, but now with advancements in modern technology, machine learning is transforming industries.
This is especially true within financial services. According to the Bank of England, two thirds of businesses within financial services use machine learning, with a third of these applications being used for a considerable share of activities in a specific business area. One area that typifies this transformation is fraud detection, which has rapidly transitioned away from traditional rules-based methods for blocking fraudulent payments, towards cutting edge machine learning, driven by data science.
The problems with traditional methods in fraud detection
Traditional methods rely on simple rules. For example, a traditional system may determine that high-value orders and orders from high-risk locations are more likely to be fraudulent. The problem with these rules is that they quickly become outdated as spending patterns or high-risk areas change. This can lead to transactions being mistakenly blocked, which can result in a loss in revenue and customer satisfaction.
To avoid this, the rules can be constantly revised and added to. However, this is time consuming and can dramatically increase the number of rules being used, which reduces the efficiency of the overall fraud detection process.
Machine Learning solves many of these problems
An increase in data causes issues for rule-based methods, but for machine learning more data generally makes for more complete models. Machine learning is a lot faster than rule-based methods which enables real-time decisions, assessing an individual customer’s behaviour as it happens. It can therefore constantly re-evaluate what normal behaviour is and as a result it becomes more accurate.
Recent research from the Association of Certified Fraud Examiners (ACFE), KPMG, and PwC has shown that criminals are using more sophisticated methods to conduct fraud. Rule-based methods find it harder and harder to identify more complicated cases of fraud. However, machine learning algorithms – both supervised and unsupervised – can help to identify the more subtle fraud patterns. Therefore, it can detect previously unseen forms of suspicious behaviour, helping companies stay on top of innovations in fraud.
The increased accuracy also reduces the human effort involved in the process, allowing staff to focus on more complex tasks for which humans are uniquely suited.
However, when machine learning models alone are used it can be difficult to understand which combination of circumstances has led to the fraud classification. Therefore, using a combination of the two models is recommended, as it allows both expert knowledge and sophisticated data science driven methods to be utilised.
How the industry is continuing to change
Companies already recognise the strengths of machine learning. Most banks have data scientists working in the fraud detection department. Many major graduate employers such as FCA, Lloyds, and PWC have data science graduate schemes relating to fraud detection. FinTech leads the way in this industry where they have been providing fraud detection beyond simply transaction monitoring. There are now examples of machine learning being used to identify suspicious trading patterns and insurance claims, among others.
Fraud detection is a classic example of how data science and machine learning have shifted an industry’s methodology quickly. The transition away from traditional methods was quick and purely performance based. Machine learning allows companies to be more accurate and therefore provide a higher quality of service to its customers.