摘要
从违约鉴别和特征选择视角出发,采用机器学习算法,对我国信息技术企业信用评级问题进行了研究。研究表明,通过SMOTE方法对非均衡样本数据进行处理,解决了样本数据类别不平衡导致预测模型存在类别偏好问题;通过Logistic-Lasso方法进行指标筛选和计算企业违约概率,并根据违约概率进行信用分级,保证了信用评级模型的精简和违约预测可靠性,改善了信用评级与违约概率不匹配问题。构建的信用评级模型采用39个指标,总体鉴别精度在98%以上,模型的可靠性和实用性优于其他常见的机器学习模型。此外,根据企业信用级别情况,分别从企业自身、投资者和监管部门角度,提出了控制风险的相应对策。
In view of default identification and feature selection,this paper uses machine learning algorithms to study the credit rating problem of information technology enterprises in China.The results shows that using SMOTE method to process imbalanced sample data solves the problem of class preference in prediction models caused by imbalanced sample data in different categories;that we use the Logistic Lasso method to select indicators and calculate the default probability of enterprises,and then to calculate credit rating,ensures the simplification of the credit rating model and the reliability of default prediction,and improves the mismatch between credit rating and default probability.The credit rating model includes 39 indicators and the identification accuracy is over 98%.The reliability and practicality of the model are superior to other common machine learning models.In addition,according to the credit level of enterprises,from the perspective of enterprises themselves,investors and regulatory authorities,the corresponding countermeasures to control risks are put forward.
作者
姜昱汐
张丽鑫
姚菲菲
Jiang Yuxi;Zhang Lixin;Yao Feifei(School of Economics and Management,Dalian Jiaotong University,Dalian 116028,Liaoning,China;Shengli Oilfield Party School(Training Center),Jinan 257097,Shandong,China)
出处
《征信》
北大核心
2024年第2期64-72,共9页
Credit Reference
基金
国家自然科学基金重点项目(71731003)
辽宁省社会科学规划基金项目(L18DTJ001)。