Zalingo Synthetic Finance — Fraudulent Transactions — 100k

Need richer signals or larger volumes? See our Focused (adds AVS/3DS/CVV, device/IP/geo consistency) and Premium Evaluation Kit (~1M rows) listings.

Dataset Features (representative)

• Core: transaction_id, ts_utc, amount, currency, channel (ecommerce | wallet | mail/phone), merchant_id, mcc, merchant_country, device_fingerprint, ip_country, user_agent
• Authorization: auth_result (approved/declined)
• Labels & Scores: fraud_label (0/1), chargeback_flag (0/1), risk_score_0_1
• Velocity (basic): velocity_txn_1h, velocity_txn_24h, velocity_txn_7d (Columns may vary slightly; see the preview file for the exact schema.)

Distribution

• Format: ZIP containing Parquet data (100k_sample.parquet), sample_100.csv (preview), and schema.json
• Volume: 100,000 rows, ~18–28 columns
• Approx Size: 3–6 MB zipped (category-dependent)
• Structure: single Parquet file (or few shards), schema stable across core fraud samples

Usage

• Baseline fraud models & rules — quick feature trials and threshold tuning
• Dashboards & KPI sandboxes — approval/decline, label rates, velocity
• Pipeline QA / MLOps — schema checks, drift probes, visualisations
• Education & demos — safe examples without cardholder data

Coverage

• Geographic: Multi-country synthetic coverage (ISO codes)
• Time Range: Recent multi-year synthetic window with weekly/seasonal patterns
• PII: None — fully synthetic; not re-identifiable

Who Can Use It

• Risk/Data Science — feature engineering, baseline benchmarks
• Payments/FinOps — loss-rate and policy experiments
• Product/Analytics — dashboards and KPI diagnostics
• Vendors/SIs — demo environments & pipeline validation

Notes / Disclaimers

• Not real cardholder data. Not for production credit decisions.
• Rates, labels, and distributions are synthetic and calibrated; they do not represent any specific issuer/acquirer/PSP.