Detecting fraud without historical data

Datasets of documented fraud cases can be instrumental in detecting anomalies and patterns in all sorts of transactions, but many times those datasets aren’t available. With fraudsters perpetrating increasingly sophisticated schemes, fraud fighters need an edge on the criminals with advanced techniques. Here, the author describes how to use machine-learning models and other advanced techniques to detect fraud when you don’t have historical data to learn from.

You might be familiar with the expression, “past behavior predicts future behavior.” When the phrase is applied in hiring, the thinking goes that a job seeker’s past successes (or failures) can help determine a candidate’s potential to thrive in an organization. The candidate’s resume or CV, behavioral interviews (interview questions designed to learn about a job candidate’s past work experiences), and recommendations from past supervisors serve as the dataset from which hiring managers can see the indicators of success — and potential red flags that suggest: “Do not hire.”