基于RF-Adaboost的企业财务困境预警研究  

Research on Financial Distress Warning of Enterprises Based on RF-Adaboost

作　　者：方逸雯 刘媛华[1] 

机构地区：[1]上海理工大学管理学院,上海

出　　处：《运筹与模糊学》2023年第1期306-314,共9页Operations Research and Fuzziology

摘　　要：针对390家上市企业样本,本文首先从获利能力、现金流量、营运能力、发展能力、偿债能力五个维度选取了13个初始财务指标,根据随机森林特征重要性筛选出了6个贡献度最大的最终财务指标。其次,本文建立了基于随机森林特征降维和Adaboost分类预测的RF-Adaboost模型,根据企业T-2年的财务指标预测其在T年是否会被特殊处理。实证结果表明RF-Adaboost模型在测试集上的分类正确率和召回率都达到80%以上。最后,为了验证RF-Adaboost模型的效果,本文还使用了Adaboost模型、LSTM神经网络、RBF-SVM、Linear-SVM、基于核密度估计的朴素贝叶斯模型进行实验,研究结果表明RF-Adaboost在所有模型中表现最好,说明了特征降维的有效性和集成算法的优越性。Aiming at the samples of 390 listed enterprises, this paper firstly selects 13 initial financial indicators from the 5 dimensions of profitability, cash flow, operating capacity, development capacity and debt paying capacity, then selects 6 final financial indicators with the greatest contribution based on the random forest’ feature importance. Secondly, this paper establishes an RF-Adaboost model based on random forest feature dimension reduction and Adaboost prediction to predict whether the enterprise will be treated specially in year T using the financial indicators data of year T-2. The empirical results show that the classification accuracy and recall rate of RF-Adaboost model on the test set are higher than 80%. Finally, in order to verify the effect of RF-Adaboost model, this paper also uses Adaboost, LSTM neural network, RBF-SVM, Linear-SVM and kernel density naive Bayes models to conduct experiments. The results show that RF-Adaboost performs best among all models, which demonstrates the effectiveness of feature dimension reduction and the superiority of integrated algorithm.

关 键 词：财务困境预警 随机森林 ADABOOST 

分 类 号：F27[经济管理—企业管理]