Case Study: Credit Card Fraud at Production Scale

The real ULB Credit Card Fraud dataset, 284,807 transactions of which only 492 are fraudulent, a five-step realistic fraud-detection workflow, and a class-weighted random forest trained on the real feature set.

~5 min total·4 quadrants of structured content

By the end of this topic, you will

State the real ULB dataset's size and fraud rate, and explain why a naive classifier's accuracy is meaningless on it
List the five steps of a realistic fraud-detection workflow in order
Train a class-weighted RandomForestClassifier on severely imbalanced data

Quadrant 1 · e-Tutorial

Video lectures and walkthroughs

Video content coming soon.

Quadrant 2 · e-Content

Articles and case studies

Case Study: Credit Card Fraud at Production Scale

A classifier that predicts legitimate for every single transaction in this real dataset is correct 99.83 percent of the time, and catches not one fraud. This is what real-world class imbalance actually looks like.

Quadrant 3 · Web Resources