FrAug: Enhanced Fraud Detection in Interbank Transfers via Augmented Account Features

In bank transfers, detecting fraud in interbank transfer is crucial for maintaining the integrity of financial systems. While graph analysis has been widely used in fraud detection systems (FDS), there have been limited attempts to apply it to interbank transfer transaction data such as that from the Housing Finance Information Network (HOFINET). This paper addresses this gap by exploring the application of the structural information inherent in transcation networks to enhance fraud detection in interbank transfers. We propose Fraud detection via feature Augmentation (FrAug), a novel approach that uses the structural information inherent in transaction networks. FrAug extracts graph centrality features and money laundering patterns from these networks, integrates them to capture inter-dependencies, and augments them to the original feature set. This method enables traditional machine learning models to capture complex patterns in graph structures, particularly forcusing on the unique characteristics of interbank transfers. We shows the effectiveness of FrAug using real-world dataset from HOFINET. Our experiments show that models enhanced with FrAug outperform their non-augmented counterparts and surpass graph neural networks (GNNs) in detecting fraudulent interbank transfers. Furthermore, our analysis provides evidence that the graph information created by FrAug are crucial for effective fraud detection.