A Deep Neural Network Based Financial Statement Fraud Detection Model: Evidence from China

Abstract

The decision-making of financial report information users largely depends on the financial data disclosed by listed companies. However, in recent years, numerous financial fraud incidents have been exposed, causing investors and stakeholders to suffer huge losses. With fraud methods of listed companies getting more and more sophisticated, the traditional financial report analysis methods have been unable to perform the detection task well. In this study, deep learning was introduced into financial statement fraud detection for the first time. Combined with 82 financial indicators, the rate of change of financial indicators and non-financial indicators, a three-layer fully connected neural network model was used to discriminate financial statement fraud of Chinese listed companies, providing a new idea for the regulatory authorities to combat fraud precisely.