Graph Attention Network Fraud Detection Based On Feature Aggregation

With the rapid development of digitalization in the financial industry, fraud detection has become an important task to ensure safe development. In traditional fraud detection tasks, training and prediction are often only based on the dimensional characteristics of a single sample, but with the development of digitization and fraud methods, these methods are often no longer applicable. In addition, there are rich information associations between users themselves, which makes a large social network graph between users. In this regard, based on the Dgraph-Fin dataset, this paper uses the relationship network formed between users to better learn the weights between different edges through feature aggregation and attention mechanism. The experimental results show that compared with the existing baselines Accuracy and validity have been improved.