首页|基于整体违约鉴别能力最优的特征选择方法研究

基于整体违约鉴别能力最优的特征选择方法研究

扫码查看
特征选择是违约判别建模的重要环节,合理的特征选择方法能够简化模型结构,提升违约鉴别效果.针对新疆维吾尔自治区内中小企业财务指标数据的特殊性,文章以整体违约鉴别能力最优为导向设计了一种新的特征选择方法,使其能够综合衡量自治区内中小企业金融借贷的风险.文章收集了新疆某银行2021年部分中小企业财务指标数据,对其进行基本的数据描述性统计和缺失值填补,并通过指标预测能力分析、多重共线性诊断及特征选择方法组合,筛选出了单个指标违约鉴别能力最强与整体违约鉴别能力最优的指标组合,有助于进一步对企业的违约状态进行判别,对企业的信用风险进行评价.通过数据对违约判别模型的有效性进行验证,其分类准确率最高达到95.85%,证明了所提模型在新疆信贷评级研究应用中具有较大的潜力.
Research on Feature Selection Method Based on Optimal Overall Default Identification Ability
Feature selection is an important part of default discriminant modeling.A reasonable feature selection method can simplify the model structure and improve the accuracy of default discrimination.In view of the particularity of financial index data of small and medium-sized enterprises(SMEs)in Xinjiang Uygur Autonomous Region,this paper designs a novel feature selection method guided by the optimal overall default discrimination ability,so that it can comprehensively measure the financial lending risk of the SMEs in the region.In this paper,we collect the SMEs'financial index data of a bank in Xinjiang in 2021,carry out basic descriptive statistics of the data and filled in the missing values.Then,through index prediction ability analysis,multicollinearity diagnosis and a combination of multiple feature selection methods,we screen out the index set with the strongest default discrimination ability of a single index and the best overall default discrimination ability.It is helpful to distinguish the SMEs'default status further and evaluate the SMEs'credit risk.The effectiveness of the discriminant model is tested on the data,and the classification accuracy reaches 95.85%,which proves that the proposed model has great potential in the research and application of credit ratings in Xinjiang.

Default discriminationfeature selectionintegrated learningdecision fusioncredit rating

王魁、成静怡、李玫璇、朱晓谦、李刚

展开 >

中国科学院大学经济与管理学院,北京 100190

东北大学工商管理学院,沈阳 110819

中国科学院大学数字经济监测预测预警与政策仿真教育部哲学社会科学实验室(培育),北京 100190

违约鉴别 特征选择 集成学习 决策融合 信贷评级

国家自然科学基金面上项目国家自然科学基金面上项目国家自然科学基金面上项目国家自然科学基金面上项目中央高校基本科研业务费专项中国科学院大学数字经济监测预测预警与政策仿真教育部哲学社会科学实验室(培育)基金

71971051723710677197120772371236

2024

系统科学与数学
中国科学院数学与系统科学研究院

系统科学与数学

CSTPCD北大核心
影响因子:0.425
ISSN:1000-0577
年,卷(期):2024.44(10)