Fraud Detection
Definition
Fraud detection examines transactions for possible fraudulent patterns and supports their review.
Fraud detection refers to identifying suspicious transactions, for example in payment data. As a specific research approach, SCARFF describes the analysis of continuous credit card transactions using machine learning and stream data processing tools.
Simply explained
Like an inspection that sets aside suspicious receipts from a stack. However, a set-aside receipt is not yet proven fraud.
Why it matters in practice
In ongoing payment streams, rare fraud cases must be distinguished from many regular transactions. SCARFF also addresses changing patterns and delayed feedback.
Example from business
Possible example: An online retailer marks suspicious payment transactions for subsequent manual review.
Limits and pitfalls
- SCARFF cites imbalanced class frequencies, changing data distributions, and delayed feedback as learning challenges. A hit must not editorially be equated with confirmed fraud evidence.
Origin & history
Carcillo and other authors submitted SCARFF as a preprint on September 26, 2017. This dates the described approach, not the overall emergence of fraud detection.
Use in mid-sized companies
- Possible application areas include screening suspicious orders, prioritizing review cases, and experiments with payment data.
Frequently asked questions
Why are delayed feedbacks a problem?
During learning, confirmation of a fraud case may only be available later.
Do fraud patterns remain unchanged?
The SCARFF work explicitly addresses non-stationary data and thus changing patterns.
Sources
- SCARFF: a Scalable Framework for Streaming Credit Card Fraud Detection with Spark · Carcillo et al. (2017): SCARFF: a Scalable Framework for Streaming Credit Card Fraud Detection with Spark
Editorially reviewed on: