Automatic detection of fraudulent behavior in networks using graph learning

• Autor(a) principal: Ronald Davi Rodrigues Pereira
• Data de Publicação: 2021
• Tipo de documento: Dissertação
• Idioma: eng
• Título da fonte: Repositório Institucional da UFMG
• Texto Completo: http://hdl.handle.net/1843/41660
• topic: Fraud Detection; Fraudulent Behavior; Graph Neural Networks; Computação – Teses; Redes neurais (Computação) – Teses; Detecção de fraude – Teses

Graph-based Neural Networks (GNNs) are recent models created for learning representations of nodes (and graphs), which have achieved promising results when detecting patterns that occur in large-scale data relating different entities. Among these patterns, financial fraud stands out for its socioeconomic relevance and for presenting particular challenges, such as the extreme imbalance between the positive (fraud) and negative (legitimate transactions) classes, and the concept drift (i.e., statistical properties of the data change over time). Since GNNs are based on message propagation, the representation of a node is strongly impacted by its neighbors and by the network's hubs, amplifying the imbalance effects. Recent works attempt to adapt undersampling and oversampling strategies for GNNs in order to mitigate this effect without, however, accounting for concept drift. In this work, we conduct experiments to evaluate existing network fraud detection techniques, considering the two previous challenges. For this, we use real datasets, complemented by synthetic data created from a new methodology introduced here. We also propose a new model framework called GMU-GNN, which performs the oversampling of graph nodes belonging to the minority class in order to improve the representativeness and expressiveness in the latent space of features interpreted by the node classification model. In new experiments carried out with 5 datasets, the GMU-GNN obtained a performance superior to the other models currently considered as state-of-the-art under the same contexts and purposes of the problem addressed here.