摘要
以科技型中小企业为研究对象,从企业的盈利能力、成长能力、运营能力、偿债能力、供应链因素五方面选取了17个影响因素,运用带有非凸惩罚的SVM模型(SCAD SVM)模型对影响中小企业的信用风险因素进行研究,并选用LassoSVM和SVM作为对比,进行变量选择和参数估计,最后对模型的准确率进行预测,得出结论:Lasso SVM方法倾向于留下一些不太重要的变量,而SCAD SVM方法通过将系数大的变量保留,系数小的直接减小为0的方式,可以选择出重要的变量,通过预测精度验证发现,SCAD SVM方法比Lasso SVM和SVM的预测精度更高.
Based on the research and development of small and medium-sized enterprises with science and technology, 17 influencing factors were selected from the aspects of profitability,growth ability, operation ability, solvency, supply chain factor and industry status. In this study, SCAD SVM model are used to research the credit risk factors of small and mediumsized enterprises, and make a comparison with the Lasso SVM and SVM in terms with variables selection and the parameters estimation. Finally, the accuracy of the model is predicted and concluded as follows: The Lasso SVM method tends to reserve more unimportant variables.The SCAD SVM method selects feature by keeping the variables with large coefficients. The SCAD SVM method is better than the SCAD SVM method. Lasso SVM and SVM have higher prediction accuracy.
作者
王少英
兰晓然
刘丽英
WANG Shao-ying;LAN Xiao-ran;LIU Li-ying(School of Mathematics and Physics,Handan College,Handan 056005,China;People's Bank of China Zhangzhou City Central Branch,Cangzhou 061000,China)
出处
《数学的实践与认识》
北大核心
2019年第3期307-312,共6页
Mathematics in Practice and Theory
基金
河北省创新能力提升计划项目18457663D
