A Data Simulation Method of Bank Fraud Transaction Based on Flow-Based Generative Model

Abstract

In order to solve the problem of category imbalance caused by the shortage of bank fraud transaction data, this paper proposes a bank fraud transaction data simulation method based on flow-based generative model. On the basis of the flow-based generative model and the real fraudulent transaction data of the bank, this method designs a generative model suitable for the bank data, and learns the distribution of the real data through the generators G and G-1. The experimental results show that mixing the generated simulated data and real business data in a certain proportion to train the fraud detection model can improve the detection effect of the model to a certain extent.