VynFi/vynfi-aml-100k

---
license: apache-2.0
task_categories:
- tabular-classification
tags:
- synthetic
- financial-data
- vynfi
- aml
- banking
- fraud-detection
size_categories:
- 100K<n<1M
---

# VynFi AML: 748K Banking Transactions with AML Labels

748,869 synthetic banking transactions. Financial services sector, 5 companies, 6 monthly periods. 59 columns including 14 pre-computed velocity features.

- 411 suspicious transactions (0.05%)
- 37,419 false positives (5% injection rate)
- 11 transaction channels (ACH, ATM, card, cash, check, mobile, online, P2P, SWIFT, wire)
- Counterparty details, velocity windows (1h/24h/7d/30d), amount z-scores

The suspicious rate is low by design. Production AML systems see comparable rates. The false-positive injection provides realistic noise for threshold calibration.

## Limitations

- Labels are generated, not from real SARs. Typology patterns are rule-based, not learned from case data.
- Velocity features are computed per-account over the synthetic timeline. They reflect the generation model's temporal logic, not real transaction behavior.
- The 100K row parameter produced 748K banking transactions due to the engine's expansion factor. This is a property of the generation, not a data error.

## Citation

```bibtex
@dataset{ivertowski_vynfi_aml_2026,
title = {VynFi AML: 748K Banking Transactions with AML Labels},
author = {Michael Ivertowski},
year = {2026},
url = {https://huggingface.co/datasets/VynFi/vynfi-aml-100k},
note = {Generated with VynFi (https://vynfi.com)}
}
```

License: Apache 2.0. Entirely synthetic.