Graph-Based Anomaly Detection for Financial Fraud Identification: Techniques and Applications

Financial networks are ubiquitous in many domains, especially in digital finance. In the financial network, there exist various entities/users such as banks, institutions, and customers, and transactions between them. Thus, it is feasible to model the financial network as graph-structured data, which records a collection of nodes (i.e., banks, institutions, and customers) and relationships among nodes (i.e., transactions). In this research, we desire to detect fraudsters and fraud teams in financial networks based on graph anomaly detection techniques. Anomalies on graphs are rare samples that appear to deviate markedly from other members, such as fraudsters (i.e. anomaly nodes) in a financial network, or money laundering networks (i.e. anomaly graphs) compared with other normal financial networks.