Fraud Risk Intelligence Platform

Real-time risk scoring over a 198,982-transaction hold-out, ranked into an analyst review queue.

XGBoost · PR-AUC 0.883 PostgreSQL + Python Sparkov

Fraud in test set

–

exposure

Alerts reviewed

–

of gated queue

Fraud caught

–

recall

Alert precision

–

loss recovered

Rank by

Review budget –

Alert budget curve recall & precision vs queue depth

Alert queue

Time	Card	Category	Amount	Prob	Exp. loss	Outcome

Fraud rate by category %

Fraud volume by week

Amount: fraud vs legit log $

Model vs rules PR