Real-Time Fraud Detection
Definition - What does Real-Time Fraud Detection mean?
Real-time fraud detection is the real-time execution of fraud-detection algorithms in order to detect fraudulent activities on credit cards and other financial payment systems. It makes use of real-time data analysis such as forensic analytics and predictive analytics to determine if an ongoing transaction is legitimate or not. Though the system is not perfect, it has reduced fraud losses in the U.S. by 70 percent since 1992, when real-time fraud detection was introduced.
Techopedia explains Real-Time Fraud Detection
Fraud detection in the simplest form is simply outlier detection, which is determining whether an event such as a purchase using a credit card occurs outside of the normal circumstances or habits of the person using it. Real-time fraud detection is just the execution of fraud detection algorithms right as the purchase is happening. The system is not perfect and a lot of false positives are captured, but this just ensures that fraud is detected immediately and possibly prevented outright. For example, a man that has exclusively been using his credit card to purchase gadgets online suddenly purchases women’s lingerie in a store from a town far away from his home. This would immediately register as an outlier occurrence because it deviates so much from the person’s purchasing habits, and depending on the credit card issuer, the transaction might be blocked or the person would get a call immediately afterwards from a representative in order to confirm whether the recent purchase was legitimate or not.
Before real-time systems made fraud detection instant, it used to be done in bulk with the results often arriving weeks or months after the purchase, which makes it difficult to track down the fraud or has allowed the culprit to commit many more fraudulent purchases before being detected and caught. This was because data used to be stored on slower disks since memory was still relatively costly. But since the cost of memory has gone down significantly since the early '90s, it has become possible to store data in-memory so that processing can happen very quickly. Real-time fraud detection can occur in as little as 40-60 milliseconds; in comparison, a human eye blink happens in 300 milliseconds. As of today, real-time fraud detection is a very common use case in the field of big data.