Fraud data science
I write about the math of catching fraud, with the data and the failure case shown.
-
Alert budgets, not AUC: judging a fraud model in the units that matter
Fraud is rare and review capacity is fixed, so judge a fraud model by the frauds in the alerts a team can open and by the money each decision moves.
-
Fraud rings and mule networks: what a graph adds, and what it cannot decide
Ring payments look normal one at a time, so the signal is in shared devices, beneficiaries and timing, and a graph score is a lead to confirm.
-
Propose, then verify: where a language model belongs in a fraud stack
A language model can propose fraud rules and explanations, but a deterministic check, separate from the model, should decide what runs in a payment flow.
-
The label you get is not the label you need
Fraud labels arrive late and only for transactions someone approved, reviewed or disputed, so training splits and evaluation have to account for that.