advancements in fraud prevention with machine learning

Machine Learning and Fraud Prevention Go Hand in Hand

Machine learning is becoming increasingly present, finding many applications in business in general and e-commerce in particular. Projections put 2022 as the year when the number of data and analytic services performed by machines will exceed 50% of the global total.

Machines Learn About Fraud

In simplest terms, machine learning refers to the ability of computers to learn through data and use that data to perform activities that would usually require human intelligence. These activities include recognizing existing and discovering new patterns of behaviour as well as predicting behaviour based on raw data.

This makes machine learning very useful for fraud detection and prevention. An important advantage of machine learning is that decisions are not only made in real time but at a speed and volume that human operators could never match during manual transaction reviews. This aspect of it turns machine learning into an extremely valuable asset in fraud prevention.


A recent example of efficient machine learning is Huawei Technologies, which has been making use of a transanalytical database that allows them to analyse credit card and mobile payments for fraudulent activity. The term “transanalytical databases” derives from a blend of "transaction" and "analysis" and are a single unified database that supports analysis of transactions in real time.



Decisions on whether a transaction will be authorized or declined are made by a machine learning model, which uses already existing fraud data in its database to identify fraudulent behaviour.

The ML model learns from each new interaction and the database itself is constantly updated with new data, allowing the model to maintain high levels of efficiency and keep the rate of false positives as low as currently possible.

Read the full article here

Machine Learning Detects Malicious Code in Milliseconds

But fraud occurs in other e-commerce venues apart from payments, such as digital advertising, which has had serious problems with ad bots. These malicious bits of code simulate human interaction with, for example, popular online videos or ads, inflating their number of views and clicks.

The only way to efficiently deal with ad bots is for the digital advertisers to analyse each interaction, which again requires a fast and accurate, real-time method to deal with what may amount to thousands of interactions per second. Once again, machine learning is a key asset, with its large and constantly updated databases which, when combined with artificial intelligence, allow the detection of data anomalies within milliseconds.


A key factor in machine learning-based fraud prevention is behavioural analytics, which allows machine learning to first understand behaviours and then anticipate them across the entire transaction.

For example, a profile can be created of every cardholder of a particular card issuer. Each profile contains information about the customer's previous card transactions and their accounts, as well as their address, requests for password resets or the creation of duplicate cards or even geographical and temporal distances between payments. Each new transaction is then analysed by machine learning and compared with the profile's previous behaviour, allowing the detection of fraudulent activity. Machine learning can also be used in behavioural analytics to predict future behaviour.


Fraudsters, of course, adapt and change and develop new tactics and approaches all the time, so it is vital that fraud prevention always be on the lookout for new threats, adapting and improving safety and security features. Adaptive analytics allow human analysts to update the fraud detection system with new information acquired during an investigation, allowing the entire system, both its human and machine learning components, to quickly and efficiently adapt to new fraud patterns, while at the same time improving the system's overall sensitivity and reducing false positives.

Mercury on the Rise

In Mercury Processing Services International’s use of machine learning, creating a system that can quickly adapt to changes has been one of the challenges, together with scalability and feature engineering, i.e. creating a set of variables that reflect a cardholder’s profile.


“We recognized the need to introduce machine learning algorithms into the fraud detection process and after using statistical models based on a simple machine learning algorithm, currently they are introducing a more sophisticated machine learning model that will be regularly re-trained and maintained internally”, explained Nataša Benčić, Senior Product Expert from Product Management and Business Consultancy, at Mercury Processing Services International.