Fraud Detection Based on Graph Neural Networks with Self-attention

With the rapid development of electronic payment, fraud cases occur frequently, and fraud detection is becoming more and more important. Traditional fraud detection model is not very good at processing information interaction between users. Furthermore, it could not handle the importance of each feature well. In order to solve this problem, this paper proposes a fraud detection model based on graph neural networks with self-attention mechanism. First of all, based on transaction data, the complex networks of interaction between user nodes and surrounding relationship nodes are reflected on the network modeling of social relationships between users. Second, by introducing graph neural networks model with self-attention mechanism based on node centrality structure characteristic index, a fraud detection model with coupling information of network characteristics and transaction characteristics is proposed. The experimental results show that this method can respond to fraud more accurately and improve the quality of judgment for the traditional fraud detection methods.