基于机器学习的康美药业财务舞弊甄别研究

Research on Financial Fraud Screening of Kangmei Pharmaceutical Based on Machine Learning

摘要: 如何有效甄别上市公司财务舞弊行为，成为业界和学界持续关注的重要议题。本研究将最近五年受到中国证监会处罚的医药生物行业A股上市公司作为样本，以康美药业为例，基于舞弊三角理论选取24个特征，采用结合SMOTE过采样技术的随机森林分类算法模型进行测试与分析。结果表明，相较于将公司简单归类为舞弊与非舞弊两类，使用多个不同的特征集建立模型或构建多个不同算法的模型进行财务舞弊甄别研究的效果更好。

Abstract: How to effectively identify financial fraud behavior of listed companies has become an important issue of continuous concern in the industry and academia. In this study, the A-share listed companies in the pharmaceutical and biological industry that have been punished by the China Securities Regulatory Commission in the past five years are taken as samples. Taking Kangmei Pharmaceutical as an example, 24 features are selected based on the fraud triangle theory, and Random forest classification algorithm model combined with SMOTE Oversampling technology is used for testing and analysis. The results indicate that compared to simply categorizing companies into fraudulent and non fraudulent categories, using multiple different feature sets to establish models or constructing models with multiple different algorithms for financial fraud screening research is more effective.