1. Dataset overview
2. Which transaction types are affected?
3. When does fraud happen?
4. What do fraudulent transactions look like?
TRANSFER and CASH_OUT transactions only.
Transaction amount
Account balance behaviour
Destination accounts receiving repeated fraud (possible mule accounts)
5. Fraud detection model (Spark MLlib)
Random Forest – confusion matrix (test set)
Feature importance
Precision / recall vs alert threshold
Lowering the threshold raises more alerts. The curve is nearly flat between 0.2 and 0.7, so the default 0.5 is a reasonable operating point.
6. Flagged transactions (test set)
7. Check a transaction
Scores a single transaction with the trained Random Forest. The first request starts Spark and takes a few seconds.
Fill with: