【作者】 朱敏；

【导师】 马汉达；

【作者基本信息】 江苏大学 ， 计算机技术（专业学位）， 2021， 硕士

【摘要】 关于中小企业的信用评级手段受到越来越多评级机构的关注,评级方法也步入了新的发展阶段——基于智能算法的评级模型,但评级模型的预测准确率极易受到数据、模型参数等的影响,不利于企业信用风险评估的过程和结果。因此,对数据和模型进行准确的分析和训练,对提高信用评级模型的性能,降低评级机构对企业的授信风险来说是具有重要意义的。为了提高针对中小企业的信用评级模型的精准性和适用性,论文使用SVM(Support Vector Machine,支持向量机)算法作为本文评级模型的基学习器,并且提出一种基于IGWO(Improved Grey Wolf Optimization,改进灰狼算法)的过采样方法并将其应用到中小企业信用评估SVM模型中去,论文主要工作如下:(1)提出IGWOSMOTE算法。针对不平衡数据的冗余特征对少数类合成过程的影响以及按固定采样倍率生成少数类样本的缺陷,提出了结合改进灰狼算法对SMOTE(Synthetic Minority Oversampling Technique,少数类样本合成技术)进行优化的方法,即IGWOSMOTE算法,详细介绍了该算法的思想和流程,最后通过实验得出该算法较传统SMOTE算法在少数类分类精度上提高了6.3个百分点。(2)提出基于IGWOSMOTE算法的SVM评级模型。首先将IGWOSMOTE算法应用于不平衡企业数据集中,建立了适用于中小企业的评级指标体系以及SVM分类模型,最终将IGWOSMOTE算法应用在SVM模型优化上,提出基于IGWOSMOTE的中小企业信用评级方法,并基于真实的企业历史数据进行模型的训练和预测,与其他模型做对比实验得出该模型具有良好的使用价值。(3)完成简易中小企业信用评级系统的设计与实现。论文简化了我国商业银行内部评级系统的全授信业务流程并分析其需求,将基于IGWOSMOTE算法的SVM评级模型应用到系统中,经过系统架构设计、功能模块分析以及数据库设计,使用计算机技术研发了简易中小企业信用评级系统为评级机构提供参考。更多还原

【Abstract】 More and more credit rating agencies pay attention to the credit rating methods of small and medium-sized enterprises(SMEs),and the rating method has entered a new stage of development--the rating model based on intelligent algorithm.However,the prediction accuracy of the rating model is easily affected by data and model parameters,which is not conducive to the process and results of enterprise credit risk assessment.Therefore,accurate analysis and training of data and model is of great significance to improve the performance of credit rating model and reduce the credit risk of rating agencies to enterprises.In order to improve the accuracy and applicability of credit rating model for small and medium-sized enterprises,this thesis uses SVM(Support Vector Machine)algorithm as the basic learner of the rating model,and proposes an over sampling method based on IGWO(Improved Grey Wolf Optimization)and applies it to the SVM model of credit rating for small and medium-sized enterprises.(1)Propose IGWOSMOTE algorithm.In view of the influence of the redundant features of unbalanced data on the process of minority class synthesis and the defect of generating minority class samples according to fixed sampling rate,this thesis proposes an improved gray wolf algorithm to optimize smote algorithm,namely IGWOSMOTE algorithm,and introduces the idea and process of the algorithm in detail.Finally,through experiments,it is concluded that the algorithm is better than the traditional smote algorithm in minority class classification accuracy 6.3 percent higher.(2)A SVM rating model based on IGWOSMOTE algorithm is proposed.Firstly,the IGWOSMOTE algorithm is applied to the imbalanced enterprise data set,and the rating index system and SVM classification model for SMEs are established.Finally,the IGWOSMOTE algorithm is applied to the optimization of SVM model,and the credit rating method for SMEs based on IGWOSMOTE is proposed.The model is trained and predicted based on the real enterprise historical data,and compared with other models It is concluded that the model has good use value.(3)A simple credit rating system for small and medium-sized enterprises is designed and implemented.This thesis simplifies the whole credit business process of internal rating system of commercial banks in China,and analyzes its demand.The SVM rating model based on IGWOSMOTE algorithm is applied to the system.After the system architecture design,functional module analysis and database design,a simple credit rating system for small and medium-sized enterprises is developed by using computer technology to provide reference for rating agencies.更多还原