Using Deep Heterogeneous Graph Learning To Identify Financial System Synthetic Identity Fraud

Abstract

One of the most advanced types of cyber-enabled financial crime is synthetic identity fraud, which costs international financial institutions billions of dollars every year. In contrast to conventional identity theft, synthetic identities get around verification systems by fusing fake personal information with authentic data. To identify such fraudulent activities in extensive financial systems, this paper suggests a novel Deep Heterogeneous Graph Learning (DHGL) framework. To find suspicious patterns that are hard to find with traditional machine learning, the suggested approach makes use of graph- based relationships between accounts, transactions, devices, and identity attributes. The framework enables robust fraud detection by capturing both structural and semantic relationships by modeling financial ecosystems as heterogeneous graphs. Our approach outperforms baseline models and conventional graph neural networks in terms of accuracy and recall, as shown by experimental results on a real-world financial transaction dataset. The goal of this research is to give financial institutions a flexible and scalable way to counteract changing fraud schemes.