ENSEMBLE LEARNING ALGORITHM - RESEARCH ANALYSIS ON THE MANAGEMENT OF FINANCIAL FRAUD AND VIOLATION IN LISTED COMPANIES

Abstract

In recent years, despite the strict "zero tolerance" crackdown on financial fraud and violations by listed companies, there has been a continued exposure of cases involving financial fraud, revenue and profit overstatement, and suspected fraud. This study first established a financial fraud index system and used the XGBoost algorithm to construct a prediction model for financial fraud and violations in listed companies. The indicators were selected and inputted into the model. A dataset was obtained for the experiments. The XGBoost algorithm was compared to two other algorithms. The receiver operating characteristic (ROC) curves showed that the XGBoost algorithm had the best prediction performance among the three algorithms. It was found that the precision of the XGBoost algorithm was 93.17%, the recall rate was 92.23%, the F1 value was 0.9270, and the area under the curve was 0.90. These results indicated better performance compared to the prediction models based on the Gradient Boosted Decision Tree (GBDT) algorithm and the Logistic algorithm. Considering the data from various evaluation indicators, it is found that the XGBoost algorithm produces the most accurate predictive effect for the financial fraud and violation prediction model.