Monkey vs Robot: How Machine Learning and AI can detect Synthetic IDs


What is Synthetic Identity Theft?

Synthetic identity theft is a type of fraud in which a criminal combines real and fake information to create a new identity.  This synthetic identity is used to open fraudulent accounts and make fraudulent purchases. Fraudsters can use multiple identities simultaneously and may even keep accounts open and active for months—even years—before the fraud is even detected.

Synthetic identity theft is now one of the most common types of identity fraud, resulting in huge losses for consumers and financial institutions. According to a report from the Federal Reserve, it is the fastest-growing financial crime in the United States costing lenders $6 billion in 2016.

Our digital identities comprise an interconnected trail of breadcrumbs spread across disconnected accounts, social networks, loyalty point programs, and library cards. In an ideal world, all of these pieces of data would connect perfectly into a rigid and robust structure: the physical address on our phone account would match exactly that on our credit history, and so on. Changes to any node would break these connections, making unsanctioned changes self-evident like a sort of multidimensional BlockChain.

In reality, the different elements of our digital identities are not perfectly correlated: when we get a new cell phone but neglect to update our membership record at the racquetball club or use our parents mailing address for packages. It is these fault planes that fraudsters exploit to create synthetic identities. They may apply for credit using all of your details except for a new mailing address to which a new card will be mailed. The federal agency’s switch to randomized Social Security, uncorrelated to date of birth, makes it easier to create synthetic identities using numbers assigned to newborns. 

If identity verification demanded strong bonds between identity molecules, no one would ever be approved for credit. However, by taking a step back and considering local connections as a whole, relative to a much larger cohort, identifiable patterns emerge where weak correlations start to look less like poor data hygiene and more like a structured attempt to circumvent the rules.  This is a task well suited to machine learning and artificial intelligence methods which can pick up on the most subtle signals in the haystack of data, learning from past mistakes and getting smarter with each iteration. 

Fraud detection and prevention is an arms race between good honest folk and those that would abuse their identities for monkey business. Attacks mutate in response to new protection measures so detection methods must be updated continuously in response, learning automatically from new data. With machine learning and AI algorithms, our robot defenders are tireless and eternally vigilant, learning from every misstep and fending off wave after wave of simian aggression.

Instnt and Continuous Identity Assurance

With Instnt Verify™, your organization can authenticate transactions from authorized account owners, effectively stopping account takeovers while aiding AML efforts. As a result, you'll gain identity assurance from account opening on Day-Zero to transactions on Day-N. Sign up today for a free demo.


About the Author

Justin is the Chief Product Officer at Instnt. He has over 20 years experience designing and building high performance distributed systems in the telecommunications, IPTV, identity verification, social media analytics, and IIoT industries. He holds patents in computer vision and collaborative filtering. He has leveraged his technical and quantitative background to deliver AI-driven solutions at Salesforce, Alcatel-Lucent, and Bell Canada, in North America, Europe, and Africa. Justin holds a degree in mechanical engineering from University of the Witwatersrand, South Africa, as well as MS in computer science and mathematics from University of New Brunswick, Canada.