Real-time Bitcoin transaction risk scoring. Live transactions are pulled from the mempool, scored by ML models for illicit activity signals, and logged to an audit trail. High-risk transactions are automatically flagged for compliance review.
This page streams live Bitcoin mempool data, scores each transaction, and displays the results below. The feed updates automatically via SSE.
Live scoring uses a mempool-features heuristic. The trained XGBoost model is available at /illicit for 166-feature input.
Lower the threshold to trigger more alerts for demo purposes.
Live mempool transactions scored in real-time.
Flagged transactions for compliance review.
| Time | TX ID | Score | Status |
|---|
Select an alert to view SHAP explanation and compliance narrative.
Review alerts as TP/FP above, then retrain the scoring model on your labels.
Score distribution stability — detects when the mempool shifts away from baseline.
Live precision, recall, F1, ROCAUC — cumulative and rolling window.
Unsupervised anomaly detection using Half-Space Trees. Learns transaction patterns and flags outliers.
3 online classifiers competing in real-time: LogisticRegression vs HoeffdingTree vs GaussianNB.