A logistic-regression fraud scorer trained on 6.36M PaySim mobile-money transactions.
The model runs in your browser — the page ships the fitted coefficients and
scaler moments as a few kilobytes of JSON, so there is no server, no cold start and nothing to wake up.
Model not loaded.Run python -m src.train then python -m src.export_web
to generate docs/model.json.
Score a transaction
—
What drove this score
The five things that moved this transaction most, strongest first. The bar shows
how much each one mattered relative to the others — it is not a percentage.