Generating Synthetic Fraud Data for Fintech Testing with fintech-fraud-sim

I Built fintech-fraud-sim: A TypeScript CLI for Synthetic Fraud Testing Data

Olamilekan Lamidi developed fintech-fraud-sim to address the difficulty of testing fraud detection systems without using sensitive production data. The tool generates synthetic users and transactions that simulate complex behavioral patterns such as account takeovers and mule activity.

Why This Matters

Fraud systems fail when tested against flat mock data because suspicious activity is defined by behavioral sequences rather than single-field anomalies. Using production data for testing introduces significant compliance risks and potential data leaks, making synthetic but realistic datasets essential for building robust risk engines and transaction monitoring services.

Key Insights

• Deterministic output via the —seed flag ensures repeatable datasets for CI/CD pipelines and regression testing suites.
• The CLI supports eight distinct fraud patterns, including mule_account and velocity_abuse, to simulate realistic adversary behavior.
• The tool generates dual-layer data, linking user metadata like KYC attempts and device counts with transaction signals such as IP country and beneficiary IDs.
• Synthetic generation avoids PII risks by excluding real names, emails, and sensitive identifiers like BVN, NIN, or bank account numbers.
• Built using TypeScript and Node.js, the package is available via NPM and supports both CSV and JSON output formats for immediate pipeline integration.

Working Examples

Basic command to generate 1,000 users with an 8% fraud rate.

npx fintech-fraud-sim generate --users 1000 --fraud-rate 0.08

Generating data with specific fraud patterns selected.

npx fintech-fraud-sim generate --users 1000 --fraud-rate 0.08 --patterns mule,account_takeover,velocity_abuse

Deterministic generation using a seed for repeatable test datasets.

npx fintech-fraud-sim generate --users 1000 --fraud-rate 0.08 --seed demo

Practical Applications

• QA Fixtures: Engineers can use seeded datasets to validate transaction monitoring rules without exposing real customer records. Pitfall: Using random, non-pattern-based mocks which fail to trigger complex risk scoring logic.
• Fraud Dashboards: Data teams can populate UI prototypes with realistic mule or account takeover scenarios for stakeholder demos. Pitfall: Manual data entry often lacks the temporal consistency required for velocity abuse simulations.

References:

• https://www.npmjs.com/package/fintech-fraud-sim
• https://dev.to/oluwatosinolamilekan/i-built-fintech-fraud-sim-a-typescript-cli-for-synthetic-fraud-testing-data-34kj