AML-GraphMiner

Explainable money-laundering detection on transaction graphs. AML-GraphMiner is an open-source Python toolkit that takes raw financial transactions and returns scored, explained alerts. It combines

  • structural-motif detectors — fan-in, fan-out, smurfing, cycles, scatter-gather and gather-scatter — and a risk-community detector (Louvain), that need no training and carry their own evidence,

  • tabular baselines (logistic regression, random forest, gradient boosting, XGBoost), supervised Graph Neural Networks (GCN, GraphSAGE, GAT) and an unsupervised graph auto-encoder, trained on leak-free, time-aware features,

  • explanations an investigator can audit (pattern evidence, occlusion, GNNExplainer, PGExplainer, Captum attributions, attention weights),

  • a web investigation interface (in French), a command line (amlgm), a REST API, a Jupyter plugin, MLflow tracking, Hugging Face Hub export and a Neo4j export, all on top of the same Python SDK.

pip install "aml-graphminer[viz,api]"
amlgm serve --demo                             # web interface on a synthetic demo: http://127.0.0.1:8000/ui/
amlgm generate data/ --accounts 1500           # synthetic IBM-AML-like data (no download needed)
amlgm scan data/Synth-Small_Trans.csv --out alerts/
amlgm train data/Synth-Small_Trans.csv --model hgb --out runs/hgb
amlgm explain runs/hgb data/Synth-Small_Trans.csv --tx 1234 --plot why.png    # any tx_id, e.g. from alerts/alerts.csv

Getting started

User guide

Interfaces and operations

Reference