Financial Crime Synthetic Data

SQLite schema + Python toolkit for generating realistic synthetic financial crime data for Transaction Monitoring.
Tables: customers (internal/external), accounts, cash_transactions (cash/ATM/card), alerts.

Purpose: Safe, PII-free sandbox to prototype TM rules, analytics and QC (KS/PSI) - with reproducible seeds.

---

Features

• Lean schema focused on TM essentials (4 tables).
• Reproducible generators (seeded) for customers, accounts, transactions and simple alerts.
• Channel-aware transaction fields (CARD/ATM/CASH with MCC, terminal/branch, etc.).
• No external dependencies (stdlib only).
• CI smoke test to build DB and verify basic integrity.

---

Data Model (v0.1)

customers (customer_id PK, is_internal) └── accounts (account_id PK, customer_id FK -> customers) └── cash_transactions (tx_id PK, account_id FK -> accounts, customer_id FK -> customers [owner], counterparty_customer_id FK -> customers [external]) alerts (alert_id PK) → references {transaction | account | customer} by (entity_type, entity_id)

Tables & key columns

• customers
  • customer_id (PK), is_internal (0/1), customer_type ('PERSON'|'ORG'), country_iso2, risk_rating ('LOW'|'MEDIUM'|'HIGH'), pep_flag (0/1), status.
• accounts
  • account_id (PK), customer_id (FK), product_type ('CHECKING'|'SAVINGS'|'CREDIT_CARD'|'PREPAID'), currency_iso3, open_date, status.
• cash_transactions
  • tx_id (PK), ts_utc, account_id (FK), customer_id (FK), counterparty_customer_id (FK), channel ('CASH'|'ATM'|'CARD'), direction ('IN'|'OUT'), amount_minor (integer, cents/grosze), currency_iso3, country_iso2, channel-specific: mcc_code, merchant_name, terminal_id, branch_id, auth_code.
• alerts
  • alert_id (PK), created_ts_utc, entity_type ('TRANSACTION'|'ACCOUNT'|'CUSTOMER'), entity_id, rule_id, score, label (0/1), typology, outcome, closed_ts_utc.

v0.1 rule of thumb: accounts belong to internal customers + counterparties are typically external.

---

Quickstart

1) Create environment

conda create -n fcc_tm python=3.11 -y

conda activate fcc_tm

2) Initialize the SQLite schema

python -m src init-db --db data/fcc_tm.sqlite

3) Generate synthetic data

python -m src generate --db data/fcc_tm.sqlite --n-customers 1500 --n-externals 1000 --n-accounts 1300 --n-transactions 200000 --seed 42

Ethics

100% synthetic data + not guidance for evasion. No affiliation with any FI or vendor.

Directory Layout

Financial-Crime-Synthetic-Data/
├─ src/
│  ├─ __main__.py         # entry: python -m src
│  ├─ cli.py              # CLI commands
│  ├─ db.py               # SQLite schema (create_db)
│  └─ gen.py              # data generators & inserters
├─ .github/
│  └─ workflows/
│     └─ ci.yml           # CI smoke test (init + generate + checks)
├─ data/                  # runtime artifacts
│  └─ fcc_tm.sqlite       # created by init-db/generate
├─ README.md
├─ pyproject.toml
├─ LICENSE