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

# VynFi AML — 748 K Banking Transactions with AML Labels

748 869 synthetic banking transactions with AML/SAR-style labels and
14 pre-computed velocity features.  Designed for tabular AML
classification benchmarks, threshold-calibration studies, and
balance-sheet / counterparty-risk analytics.

Generated with **DataSynth** (banking module, KYC/AML typologies) ·
[GitHub](https://github.com/mivertowski/SyntheticData) ·
[Companion paper (SSRN)](https://ssrn.com/abstract=6538639).

> **Provenance note.**  This dataset was last refreshed under
> DataSynth v5.0.  The v5.6 → v5.9 release line ships fixes
> centred on the journal-entry / accounting-network generators;
> the banking module producing this dataset was untouched, so the
> data here is unaffected by those changes and a regeneration is
> not in scope for the v5.9.0 fleet refresh.

## What's included

| Config | Rows | Columns | What it is |
|---|---|---|---|
| (default) | 748,869 | 59 | One row per banking transaction with counterparty, channel, amount, FX, velocity-window aggregates, and AML labels (`is_suspicious`, `typology`, `false_positive`). |

## Schema highlights

* **Channels (11)** — `ACH`, `ATM`, `card`, `cash`, `check`, `mobile`, `online`, `P2P`, `SWIFT`, `wire`, plus an `internal` transfer category.
* **Counterparty** — `counterparty_id`, `counterparty_name`, `counterparty_country`, `counterparty_account_type`.
* **Velocity windows** — 1 h, 24 h, 7 d, 30 d aggregates of incoming + outgoing amounts and counts, plus `amount_zscore_*` and `velocity_zscore_*` features.
* **AML labels** — 411 suspicious transactions (0.05 %), 37 419 false-positives (5 %).
* **Typologies covered** — structuring / smurfing, layering, mule activity, round-tripping, fraud, market spoofing.

## Quick start

```python
from datasets import load_dataset

txns = load_dataset("VynFi/vynfi-aml-100k", split="train")
print(txns.features)

# Suspicious-only slice
suspicious = txns.filter(lambda r: r["is_suspicious"])
```

## Generation

| | |
|---|---|
| **DataSynth release** | v5.0 (banking module — unchanged in v5.6–v5.9) |
| **Industry** | Financial services |
| **Period** | 6 monthly periods |
| **Companies** | 5 |
| **Suspicious rate** | 0.05 % (matches production AML transaction rates) |
| **False-positive injection** | 5 % (for threshold calibration) |
| **Reproducibility** | Determined by the embedded ChaCha8 seed; regenerable from a future v5.9-aware banking config (not yet pinned in `configs/examples/hf/`). |

## 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 generator's temporal model, not real
  transaction behaviour.
* The "100k" in the dataset name refers to the row-count parameter
  passed to the engine; the banking module's expansion factor
  produces 748 869 rows.  This is a property of the generation
  model, not a data error.

## License

Apache-2.0.

## Citation

```bibtex
@misc{ivertowski2026datasynth,
  author       = {Ivertowski, Michael},
  title        = {{DataSynth}: Reference Knowledge Graphs for Enterprise
                  Audit Analytics through Synthetic Data Generation
                  with Provable Statistical Properties},
  year         = {2026},
  month        = {April},
  howpublished = {SSRN Working Paper},
  url          = {https://ssrn.com/abstract=6538639}
}
```