Training set
Generate 3 000 fictional transactions; ~6% are anomalies (large amount + odd hour + new merchant + far from home).
Live · Banking & Financial Services
Fraud-risk scoring
Enter a synthetic transaction below. An IsolationForest trained on 3 000 fictional transactions returns a 0–1 risk score with the features driving the decision — exactly the structured artefact your auditor needs.
Try a scenario
Risk score
Submit a transaction to score.
How it works
IsolationForest
Unsupervised model — trees isolate points by random feature splits. Anomalies isolate faster, get higher scores.
Score calibration
Raw decision_function output is normalised to a 0–1 range against the training distribution.
Feature drivers
For each new transaction, compute z-scores against the training-distribution per feature. The squared z² gives the contribution share.
Verdict thresholds
< 0.4 low-risk · 0.4–0.7 review · > 0.7 high-risk. Thresholds tunable in production to match the team's false-positive budget.
Production swap
Replace with a gradient-boosted ensemble trained on labelled fraud cases, calibrated with isotonic regression, and exported with full feature attributions per decision.
Want this on your real transactions?
We build production fraud-decisioning systems with your card / transfer streams, calibrated to your false-positive tolerance, integrated into your existing case-management tooling, and documented for model risk management.
Ready to start
Turn one AI use case into measurable production value.
Book a 30-minute consultation. We will walk through the use case, sketch the value case, and tell you honestly whether we can help.