This system looks at a credit card transaction and decides in milliseconds whether it is likely to be fraudulent or legitimate — the same way your bank automatically flags suspicious purchases on your card.
The system was trained on a real dataset of credit card transactions — each one labeled as either legitimate or fraudulent. It learned the patterns that separate normal purchases from suspicious ones.
It doesn't just look at the transaction amount. It considers behavioral patterns — how many transactions happened recently, what time of day it is, whether the amount is unusually large compared to recent history.
For every new transaction, the system returns a label (Legit or Fraud) and a probability score — how confident it is. A score of 4% means very likely legitimate. A score of 95% means very likely fraud.
Here is what happens when a transaction is submitted to the API:
A score of 0.98 out of 1.0 means that if you showed the model a real fraud case and a legitimate transaction side by side, it would correctly identify the fraudulent one 98% of the time. It catches 83% of all fraud cases in the test set.