基于XGBoost-LR模型的企业财务危机预测与研究  

Enterprise Financial Crisis Prediction and Research Based on XGBoost-LR Algorithm

作　　者：王舒梵 姜新盈 严涛 WANG Shufan;JIANG Xinying;YAN Tao(Shanghai University of Engineering Science,School of Mathematics,Physics and Statistics,Shanghai 201600)

机构地区：[1]上海工程技术大学数理与统计学院,上海201600

出　　处：《软件》2022年第4期13-17,共5页Software

基　　金：全国统计科学研究项目(2018LY16)。

摘　　要：上市公司年报中的统计性数据是上市公司发展情况的重要体现,通过对上市公司经营情况数据的挖掘与分析可以提高对其财务风险的预测能力。本文从上市公司的财务数据中选取了36个财务指标,并把指标划分成五个维度进行分析(依次是偿债能力、盈利能力、营运能力、发展能力和现金流能力)[1],分别使用Logistic回归算法、决策树算法、支持向量机算法、极限梯度提升决策树(XGBoost)等算法建立了企业经典财务预警模型,并提出了一种基于XGBoost-LR算法的融合模型来进行财务危机预警,该模型首先运用用XGBoost算法自动进行了特征组合和离散化,然后将新构造出的特征向量运用在LR模型上,通过LR模型来进行分类预测。通过模型对比表明,该算法在预测企业财务危机上具有较高的准确率,可为企业日常生产经营活动提供一定的指导意义。The statistical data in the annual report of listed companies is an important reflection of the development of listed companies. Through the mining and analysis of the operation data of listed companies, the predictive ability of their financial risks can be improved. This paper selects 36 financial indicators from the financial data of listed companies and divides them into five dimensions for analysis(debt paying ability, profitability, operation ability, development ability and cash flow ability in turn). The classical financial warning model is established by using Logistic regression algorithm, decision tree algorithm, support vector machine algorithm, limit gradient enhancement decision tree(XGBoost) and other algorithms, and a fusion model based on XGBoost-LR algorithm is proposed to carry out financial crisis warning. In this model, XGBoost algorithm is used to automate feature combination and discretization, and then the newly constructed feature vector is applied to LR model, and classification prediction is carried out through LR model. The model comparison shows that the algorithm has a high accuracy in predicting the financial crisis of enterprises, which can provide some guidance for the daily production and operation of enterprises.

关 键 词：财务危机 财务危机预警模型 不平衡数据 XGBoost-LR融合模型 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程] F27[自动化与计算机技术—控制科学与工程]