Part 3: Building AI for fraud detection
Financial fraud is an arms race. Fraudsters adapt, industrialize, and collaborate—and the AI systems built to stop them must do the same. This part equips you with the full spectrum of detection techniques, from scoring individual transactions to uncovering organized criminal networks hidden across millions of data points.
Chapter 7 surveys the landscape of modern financial fraud—payment fraud, identity theft, collusive schemes, and more—and maps these threats onto the 4-Layer Framework, establishing the architectural principles for real-time, adversarial detection systems. In chapter 8, you move from concept to code: preparing highly imbalanced fraud data, building supervised models with LightGBM to catch known patterns, and deploying unsupervised anomaly detection to flag entirely novel threats. Chapter 9 then takes you beyond isolated transactions into the world of graph analytics, where you’ll use NetworkX and Graph Neural Networks to mathematically expose fraud rings, money laundering cells, and coordinated schemes that no single-transaction model could ever detect.
When you’ve completed this part, you’ll understand how to layer supervised, unsupervised, and graph-based methods into a multi-tiered fraud defense system built for the realities of production.