Trained with PyTorch on synthetic transactions in a Chalk training run. It forecasts each customer's next transaction and scores every transaction for fraud.
Fraud precision by epoch (held-out customers)
Customerred bar = model's fraud score · border = actual fraud