Design a fraud detection system that scores every payment authorization on the processor's authorization path. For each transaction, the system must quickly determine whether it looks like legitimate activity, stolen-card use, account takeover, or another form of fraud without unnecessarily rejecting legitimate customers.
The ML task is binary classification with calibrated output. Given transaction, account, device, merchant, historical, and graph signals, the system…
