基于XGBoost的企业倒闭风险预测

XGBoost-based enterprise failure risk prediction

企业经营风险预测对于判断企业的经营状况,指导企业的进一步行为具有实际意义。然而,对于大量的小微型企业来说,一般很难获得这些企业的真实财务信息,也缺乏全面的信用信息供参考。文章以企业主体在多方面留下的行为足迹信息构建训练集,基于不同的足迹行为数据分别使用eXtreme Gradient Boosting(XGBoost)算法构建预测模型,并使用线性加权融合多个模型,以企业在未来两年是否会退出市场为目标变量进行预测。结果表明,在现有数据的基础上,该模型可以有效预测企业的经营风险,相比于传统的方法,精度更高。

The prediction of business risk has practical significance for judging the business status of enterprises and guiding the further actions of enterprises.However,for a large number of small and micro enterprises,it is generally difficult to obtain the true financial information of these companies,and there is no comprehensive credit information for reference.This paper builds a training set based on the behavioral footprint information left by the company's main body,and uses the eXtreme Gradient Boosting(XGBoost)algorithm to build prediction models based on different footprint behavior data,and uses linear weighted fusion to fuse multiple models to the enterprise in the next two.Whether the year will exit the market and predict the target variable.The results show that based on the existing data,the model can effectively predict the business risk of the company,and the accuracy is higher than traditional methods.