fraud prevention using ai and machine learning in practice
Nets and KPMG new whitepaper, Fighting fraud with a Model of Models reviews how financial institutions can harness advances in AI and machine learning to combat card fraud.
Fighting Fraud with a Model of Models
Fighting Fraud with a Model of Models, the whitepaper made by Nets and KPMG, explains how utilising human expertise in combination with artificial intelligence (AI) and machine learning (ML) technologies can significantly increase the accuracy of fraud prevention services.
Moreover, the paper explores the theoretical approach behind Nets Fraud Ensemble, an AI-powered anti-fraud engine developed in collaboration with KPMG, which can reduce fraudulent transactions by up to 40% on top of existing AI fraud prevention measures, for the benefits of banks, merchants and cardholders, as well as society in general.
The paper also brings a very comprehensive analysis of today's fraud landscape, as well as some very valuable advice on leveraging historic data, using fraudsters' limitations against them, threshold optimisation, model explainability.
“Until now, the use of true machine learning to fight payment card fraud has been limited. With Machine Learning and AI we can find patterns in the data which are too complex for the human brain to identify, analyse and then act on them to prevent fraudulent transactions.
Sune Gabelgård, Head of Fraud, Intelligence & Research, Nets
Nets Fraud Ensemble
Nets Fraud Ensemble is an AI-powered anti-fraud engine that improves real time fraud prevention in an ever-changing landscape. By deploying true machine learning (i.e. a system that automatically identifies and reacts to existing and new fraud patterns), it represents a significant step forward from the rules-based models that are currently in use across the international banking industry.
The solution resulted in an immediate fraud reduction of 25% and an estimated 40% long-term potential.
The ‘brain’ of Nets Fraud Ensemble consists of multiple models working together to analyse each individual transaction within ten milliseconds – the time frame in which a transaction can be blocked. The solution learns automatically from patterns observed in the data and adjusts accordingly.
This means that the longer historic data series available to it, the more fraudulent transactions are blocked, and the fewer false positives are raised.