Design of XGBoost prediction model for financial operation fraud of listed companies

To resolve the issue of untimely discovery of fraud phenomena in the supervision process of listed companies’ financial fraud, this research studies the establishment of a financial operation fraud prediction model for companies based on XGBoost and introduces regularization items and column sampling to enhance the robustness. Meanwhile, the data processing strategy of the model is designed according to the characteristics of the fraud data of listed companies, and the gray samples in the data samples are eliminated. Finally, the research uses the financial fraud prediction test method to test the model. The results indicated that the AUC of the XGBoost model designed in the study was 0.91, which was the maximum value of AUC in the comparison model. And its prediction effect reached the greatest level. The XGBoost prediction model designed by the research had a KS value of 0.65 in the prediction, which was the best value among the comparison models. This value was within the range of the KS value of the ideal model and can well distinguish positive and negative samples. The XGBoost listed company financial operation fraud prediction model designed by the research can effectively and dynamically predict financial fraud, laying a foundation for the establishment of a comprehensive capital market supervision system.