A Study of Identification of Corporate Financial Fraud Using Neural Network Algorithms in an Information-based Environment

Abstract

This paper provides a brief overview of corporate financial fraud behavior and the initial feature indicators utilized for detecting financial fraud. Principal Component Analysis (PCA) was employed to refine these feature indicators. Subsequently, the Back-Propagation Neural Network (BPNN) algorithm was applied for identification. Simulation experiments were conducted to test the BPNN algorithm's parameters. Additionally, a comparative analysis was conducted to compare the BPNN algorithm with the decision tree and Support Vector Machine (SVM) algorithms. The results demonstrated that PCA effectively reduced the initial set of 30 indicators to 20, retaining 90.64% of the essential information. The optimal configuration for the BPNN algorithm was seven hidden nodes and the application of the ReLU activation function. Furthermore, the BPNN algorithm outperformed the decision tree and SVM algorithms in the context of financial fraud recognition.