ziadatalabs/FreeSyntheticFraudRings

Free Synthetic Fraud Rings (2M Accounts / 30M Transactions)

A two-table synthetic financial network built for graph-based fraud detection and anti-money-laundering research: 2,000,000 accounts and 30,000,000 transactions between them, with a defined set of coordinated fraud rings hidden inside an otherwise normal transaction graph.

Most fraud datasets label individual transactions in isolation. This one labels structure. A set of fraud rings operates as coordinated money-laundering networks: funds enter from victim accounts, hop through two to four layers of mule accounts, and consolidate to a ring cash-out account. Mules in a ring share a device and signed up in a burst. The ring_id field is the ground-truth label on both tables, so you can train and evaluate against the true network — not just guess.

Two properties make this hard in the right way. First, the money balances: what a ring pulls in from victims routes through its mules and out to cash-out, minus a retained cut — per-ring totals reconcile. Second, it's deliberately subtle: mule transaction amounts sit inside the normal amount distribution (roughly three-quarters fall within the legit p10–p90 band), fraud is about 1.3% of all transactions, and only a fraction of ring accounts carry an overt risk flag. The signal lives in the graph — shared devices, layered flow, timing — not in any single oversized transaction, so a trivial amount threshold won't find the rings but a graph model can.

Schema

fraud_accounts.parquet — 2,000,000 rows

Column	Type	Description
account_id	string	Unique account identifier (primary key)
account_type	string	personal, business, mule, or cashout
signup_date	string	Account signup date (YYYY-MM-DD)
home_region	string	Account home region
device_id	string	Device fingerprint — shared across a ring's members
account_age_days	int	Age of the account in days
risk_flag	bool	Whether the account carries an overt risk flag
ring_id	string	Ground-truth ring label; empty for legit accounts

fraud_transactions.parquet — 30,000,000 rows

Column	Type	Description
txn_id	string	Unique transaction identifier
src_account	string	Source account (foreign key to accounts)
dst_account	string	Destination account (foreign key to accounts)
amount	float	Transaction amount
timestamp	string	Transaction time (YYYY-MM-DD HH:MM:SS)
channel	string	card, ach, wire, p2p, mobile, or crypto_onramp
is_fraud	bool	Whether the transaction is part of a ring flow
ring_id	string	Ground-truth ring label; empty for legit transactions

Every src_account and dst_account resolves to a row in the accounts table.

Format

Two Parquet files, Snappy-compressed.

Quick start

import pandas as pd
accounts = pd.read_parquet("fraud_accounts.parquet")
txns = pd.read_parquet("fraud_transactions.parquet")

import datasets
ds = datasets.load_dataset("ziadatalabs/FreeSyntheticFraudRings")

import duckdb
# rebuild one ring's money flow
duckdb.sql("""
  SELECT src_account, dst_account, amount, channel
  FROM 'fraud_transactions.parquet'
  WHERE ring_id = 'RING-00000'
  ORDER BY timestamp
""").show()

Notes

All data is synthetic — no real accounts, customers, or financial records of any kind. Ring structures, amounts, and labels are generated. Built for research, benchmarking, and prototyping fraud- and AML-detection methods.

License & Usage

CC-BY-NC-4.0. Free for research, learning, and non-commercial use.

Want more free datasets? Hit the ❤️ and follow. And we take requests — tell us what synthetic data you need, and we'll build it.

Zia Data Labs · zia.data.team@protonmail.com