ARTIFICIAL INTELLIGENCE FOR FRAUD DETECTION
Volume II in the Fraud Detection series — the applied companion to Mathematical Theory of Fraud Detection.
Volume I built the mathematical foundation: a lead generation platform as a dynamic probabilistic system, with fraud as a function of a hidden quality state inferred from noisy evidence. Artificial Intelligence for Fraud Detection picks up where the theory ends and shows how to build, explain, and operate the systems that put it into production.
This volume covers the full arc from model to machine: explainable AI (SHAP, LIME, counterfactual reasoning) for defending automated decisions to regulators and analysts alike; large language models as investigative assistants for log analysis, alert summarization, and knowledge-graph-driven case work; and autonomous, agentic detection systems that monitor, decide, and act with a human in (or on) the loop.
From there, the book moves into engineering: real-time scoring pipelines built as online Bayesian updating, the distributed infrastructure that makes it possible — Kafka, Spark, ClickHouse, Redis — and the privacy, security, and compliance obligations (GDPR, differential privacy, federated learning) that come with deploying AI against real user data.
Inside this book, you'll find:
• Explainable AI methods — SHAP, LIME, feature attribution, and counterfactual explanations — built for regulatory and operational scrutiny
• LLM-powered investigation: log analysis, alert triage, analyst assistants, and fraud knowledge graphs
• Agent architectures for autonomous, self-learning fraud detection, including multi-agent coordination and human-AI teaming
• Production-grade real-time pipelines: feature stores, streaming inference, and distributed serving infrastructure
• Privacy-preserving and compliant AI: differential privacy, federated learning, and encryption for sensitive marketplace data
Written by Professor Yuri Grant and software specialist and tech entrepreneur Maksym Holovchenko — combining academic research in probabilistic modeling with decades of production experience building high-throughput lead distribution platforms — this book is for the data scientists, ML engineers, and fraud and risk teams who have to make the theory work at scale. Read alongside Volume I for the complete picture, or on its own as a practitioner's guide to AI-driven fraud defense.