Fraud/Synthetic
Fraud scenarios
Ground truth joined to the bank and card transactions. Each transaction has is_fraud and, when it is fraud, a fraud type: card testing, card-not-present fraud, account takeover, or authorised push-payment scam. The same table also marks AML: structuring, mule networks, and rapid pass-through. Difficulty is easy, medium, or hard, and controls how close a scenario sits to ordinary spending. A holdout window separates earlier activity from a later test period. These are not a second invented population. They are the labels the bank simulation writes. No real cardholder data is used.
- Volume
- 10,000customer sample
- Formats
- CSV · Parquet · JSONL
- Label
- Synthetic
Purchase
Sold on Datarade. This site does not take payment.
Placeholder — listing URL not set
Schema
Column preview for the launch specification. Names and types describe the file. They are not a sample of rows.
- txn_idstring
Transaction this label belongs to.
- is_fraudboolean
Whether the transaction is part of a fraud scenario.
- fraud_typestring
card_testing, cnp_fraud, account_takeover, or app_scam. Empty when is_fraud is false.
- is_amlboolean
Whether the transaction is part of an AML scenario.
- aml_typologystring
structuring, mule_network, or rapid_pass_through. Empty when is_aml is false.
- difficultystring
easy, medium, or hard.
- scenario_idstring
The injected scenario, when this row is part of one.
Use
- Train a fraud classifier on labeled transactions
- Test a model on the holdout window
Generation
Fraud and AML labels generated with the bank simulation. Not copied from real fraud cases.
- Every dataset is 100% synthetic. It is not copied from real customers, cardholders, or patients.
- Each product is labeled as synthetic in the title, the description, and the file metadata.
- Delivery includes a generation note: the methods and algorithms used, the schema, and known limitations.