Fraud detection is a typical usecase that requires graphs. It’s all about relationships between identifiers, and about quickly detecting that they are being shared. Shapes and sizes of fraud rings can vary considerably. Neo4j’s Cypher language facilitates the creation of pattern recognition capabilities to help detect fraud rings as they emerge, and when combined with our InterActor, provide a vizualisation of the fraud ring itself. At the same time suspected fraudulent identities and victims are spotted.
For this demo, 400 valid identities are created, each identity has a name, an address, a social security number (ssn) and a phonenumber. In addition, 64 fake identities are created based on the cartesian product of a set of 4 randomly chosen addresses,ssn and phone numbers from the valid identities. So, in total we have 464 identities.
As long as these identities don’t have opened accounts in their names, it’s not possible to see which identities are fraudulent and which are not. Even when there is just a single fraudulent identity that opens an account, it’s difficult. But when we detect that account-holders start to share addresses, ssn and phone, the pattern emerges rapidly. In this demo, we create individual random accounts, and once we suspect fraud, the InterActor application identifies identities that may be fraudulent, as well as identities that may be victims. The latter are those that share only one of their valid identifiers (ssn, address or phone) with other account-holders. Note in the video that an identity that is initially identified as a victim, later on appears to be member of the fraud ring.
Docker
Cloud