Spaces:
Sleeping
Sleeping
File size: 1,730 Bytes
a585e24 ee9eb6a a585e24 ee9eb6a a585e24 ee9eb6a a585e24 ee9eb6a | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 | ---
title: VynFi Fraud-GNN Demo
emoji: 🛡️
colorFrom: red
colorTo: indigo
sdk: gradio
sdk_version: 5.5.0
python_version: '3.11'
app_file: app.py
pinned: true
license: apache-2.0
short_description: GraphSAGE fraud + GAE anomaly on synthetic JE network
tags:
- vynfi
- graph-neural-network
- fraud-detection
- anomaly-detection
- synthetic-data
---
# 🛡️ VynFi Fraud-GNN Demo
Interactive inference Space for the
[`VynFi/je-fraud-gnn`](https://huggingface.co/VynFi/je-fraud-gnn)
model bundle.
## Three tabs
* **Edge fraud predictor** — pick a curated sample (clear fraud / clear
normal / borderline) or build your own edge from any of the 499 GL
accounts in the published COA. Returns fraud probability + anomaly MSE.
* **Node anomaly explorer** — top-K accounts ranked by GAE
reconstruction error on a 5,000-edge sample; surfaces accounts whose
attribute patterns don't fit the structural prior.
* **Live evaluation** — sample N edges from
[`VynFi/vynfi-journal-entries-1m`](https://huggingface.co/datasets/VynFi/vynfi-journal-entries-1m),
run the classifier, render confusion matrix + ROC against ground truth.
## Tech
* Gradio + torch-geometric + pandas + matplotlib
* Loads model bundle from `VynFi/je-fraud-gnn` at cold-start (cached after).
* Loads dataset slices from `VynFi/vynfi-journal-entries-1m` on demand.
## Source
* [Engine repo (`spaces/fraud-gnn-demo/`)](https://github.com/mivertowski/SyntheticData/tree/main/spaces/fraud-gnn-demo)
* [Model card](https://huggingface.co/VynFi/je-fraud-gnn) — full training details, metrics, and honest discussion of where GNN helps vs LR baseline.
* [Companion paper (SSRN)](https://ssrn.com/abstract=6538639)
## License
Apache-2.0.
|