Improving Fraud Ring Detection Using Machine Learning to Reduce Credit Card Application Fraud

Download this whitepaper to explore how applying different characteristics of fraud rings can improve machine learning (ML) solutions to more effectively mitigate fraud in credit card account originations. Analyzing credit card application data with verified fraud feedback and the TransUnion^® identity graph (which approximates credit-active identities across the US), we determined 10% of all credit card applications belong to the top 1% of fraud rings — which are responsible for 25% of total fraudulent applications.

How can machine learning models improve fraud ring detection?

We define a fraud ring as a cluster of at least two identities whose applications were fraudulent at least 50% of the time. Out of 40,000 groups studied, 400 (1%) were identified as fraud rings. In three case studies that show varying fraud density (low, high and very high), we illustrate how network information within fraud rings provide valuable context for fortifying fraud detection using machine learning models.

Submit the form to download the whitepaper