A financial statement fraud detection model based on hybrid data mining methods

Financial statement fraud has been a difficult problem for both the public and government regulators, so various data mining methods have been used for financial statement fraud detection to provide decision support for stakeholders. The purpose of this study is to propose an optimized financial fraud detection model combining feature selection and machine learning classification. The study indicated that random forest outperformed the other four methods. As to two feature selection methods, Xgboost performed better. And according to our research, 2 or 5 variables are more acceptable for models in this paper.