Glossary · E-commerce ML

Real-Time Fraud Detection

also: transaction fraud scoring · payment fraud ML

Definition

Real-time fraud detection scores checkout transactions within latency budgets of 50–200 ms to decide allow, challenge, or block. Production systems combine gradient boosting (feature-rich), graph features (linked-device, shared-card), and autoencoder-based anomaly scoring under extreme class imbalance and adversarial adaptation.

Payment fraud is adversarial and highly imbalanced (fraud rates of 0.1–1%). Production systems layer models: a fast gradient-boosted tree for most transactions, autoencoders for novelty detection, and graph features that catch device/card/IP ring attacks. Decision thresholds balance false positives (legitimate cart abandonment) against false negatives (chargebacks and losses). Stripe Radar, Shopify Kount, and internal systems at large marketplaces follow this general architecture.

Essays on this concept