How machine learning can help FIs prevent fraud

Data analytics is used in many businesses to derive conclusions from information across a range of potential sources. This may be customer details, market research, purchase information or a multitude of other data inputs.

Organizations use advanced modelling techniques and sophisticated software to deliver actionable insights that influence decision-making and strategic positioning. Many financial institutions (FIs), meanwhile, are fairly well established in their use of data to identify and prevent fraud.

Machine learning has been particularly useful for this task, so let’s explore how it works and the ways in which FIs can use machine learning technology to optimize fraud detection.

Machine learning: the basicsĀ 

At its simplest, machine learning refers to computers that have the ability to learn and develop without relying on specific programming. They become better at performing a chosen task over time after continual exposure to new data.

Machine learning therefore enables systems to recognize patterns, make predictions and exhibit other ‘intelligent’ behaviors once they have digested huge datasets, and then to continually evolve their learnings as information is updated. The concept has been around for decades, but interest in the field has grown significantly in recent years due to increases in processing power and the spread of data-generating devices, among other factors.

Two broad categories of machine learning exist: supervised and unsupervised. The former relies on the computer being fed a predefined set of ‘correct’ inputs and outputs that help the machine spot errors in new datasets and adjust models accordingly. The latter provides no such guidance and essentially allows the computer to identify hidden patterns in data on its own, which the machine then groups into categories based on similar traits.

Machine learning and FIs

It is not difficult to see how this type of technology could be useful for fraud identification in the finance industry. Banks can have tens of millions of customers worldwide, which means a phenomenal number of transactions and new account openings each day.

Figures from Javelin Research, published in February 2016, revealed that 13.1 million people in the US were the victims of fraud last year. The amount of money stolen from these individuals totaled $15 billion, with new account fraud doubling during this time.

Clearly, manual processes can only go so far in detecting fraudulent activity, which means accurate and reliable automated solutions are required to keep both customers and FIs safe from criminals.

Machine learning means organizations can load computers with historical data containing fraudulent cases, enabling the system to process and analyze future transactions in real-time.

Benefiting from machine learning

The obvious advantages to machine learning are that fraud detection becomes quicker and more accurate. However, the financial benefits are just the tip of the iceberg. Fraud identification and prevention must be balanced with customer convenience to maintain satisfaction levels – incorrectly blocking card payments due to suspected fraudulent behaviour (known as false positives) is often frustrating for legitimate consumers.

Machine learning and other data analytics techniques can therefore help businesses avoid the reputational damage that typically accompanies failed fraud identification processes.

As FIs become more adept at collecting, handling and analizing data, it’s likely that they will increasingly use machine-learning technology to drive fraud prevention within their operations.

Written by Dena Hamilton

Dena is NCR's Director of Enterprise Fraud & Security Software Solutions. She specializes in fraud, risk, compliance and security, with over 35 years of experience in the financial services space. Her focus is the development and deployment of enterprise financial crime solutions optimized in prevention, detection and back office efficiency.

Read more articles from Dena Hamilton