基于决策树的上市公司风险分类与预测
Risk Classification and Forecast of Listed Companies Based on Decision Tree

上市公司财务造假、违规担保等违规犯罪行为屡见不鲜，对上市公司合理的分类评级对维持金融市场秩序具有重要意义。本研究构建了一套指标类型为基础性和触发性的风险分类评级指标体系，使用触发性指标应用决策树算法对收集到的上市公司进行风险分类，使用基础性指标构建决策树模型用于风险预测，并且针对错分代价的不平衡等问题，对分类为正常上市公司的样本使用成本敏感决策树进行二次分类。结果表明，用于分类的模型准确率达到100%，用于预测的模型训练集准确度为92.7%，测试集准确度为80%，成本敏感决策树的二次分类，将有风险上市公司分类准确率提升至100%，整体准确率由91.7%提高到96.7%。
Financial fraud, illegal guarantee and other illegal crimes of listed companies are common, and a reasonable classification and rating of listed companies is of great significance to maintaining the order of financial market. In this study, a set of risk classification and rating index system with basic and trigger indicators is constructed. The trigger indicators are used to classify the risks of the listed companies by decision tree algorithm, and the classification accuracy of training set and test set is 100%. Then, the decision tree model is built by using basic indicators for prediction. In order to improve the prediction accuracy, the relationship between the model and the sector where listed companies are located is discussed. In view of the imbalance of misclassification cost, the samples classified as normal listed companies are classified twice by using cost-sensitive decision tree. The results show that the accuracy of the model training set used for prediction is 92.7%, the accuracy of the test set is 80%, the segmentation of listed companies can improve the accuracy of the model, and the secondary classification of cost-sensitive decision tree can improve the classification accuracy of risky listed companies to 100% and the overall accuracy from 91.7% to 96.7%.