Synthetic Identity Detection using Inductive Graph Convolutional Networks

Synthetic identities (SIDs) frequently blend factual and fabricated personal information to perpetrate fraud. Throughout the pandemic, both the number and scale of data breaches have increased, which adds to the growth in SID- based fraud. The standard procedure for finding SIDs involves extracting the dense portion of the graph that is poorly linked to other nodes. Current methods for detecting dense regions focus on maximizing density, while others aim to reduce the number of missing edges. However, existing algorithms rely on obsolete graph mining techniques and cannot differentiate between malicious and benign newly arriving nodes. Hence, we present a Graph Convolutional Network (GCN)-based SID detection approach that may identify fraudsters or groups of fraudsters who are not typical. Our method can tell if a freshly arriving node is a fraudster.