Graph-Based Approaches for Real-Time Fraud Detection Systems

This book explores advanced graph-based methods for real-time fraud detection systems, focusing on how complex relationships between entities such as users, accounts and devices can be modeled as graphs. By leveraging Graph Neural Networks (GNNs) and temporal modeling, it demonstrates how modern AI techniques can detect sophisticated, coordinated fraud patterns that traditional rule-based and statistical systems often fail to identify.It traces the evolution of fraud detection approaches and highlights the limitations of legacy systems in handling scale, speed, and evolving attack strategies. The book further examines heterogeneous and temporal graph structures, as well as key real-time challenges including scalability, low latency, concept drift, explainability, ethics, and continuous learning.A hybrid framework is proposed, combining offline training with online inference to ensure both robustness and adaptability in dynamic environments. Written by Sara Kaya (Somayeh Babaeitarkami) and Dr. Ali Kaya, the book serves as a practical and strategic guide for researchers, data scientists, and practitioners building next-generation, transparent, and adaptive fraud detection systems.