基于FWGKNN算法的个人违约率的应用  

Application of Probability of Default Evaluation Based on the FWGKNN Algorithm

作　　者：葛传明 GE Chuan-ming(School of Management Science and Engineering,Anhui University of Technology,Anhui 243002)

机构地区：[1]安徽工业大学管理科学与工程学院,安徽243002

出　　处：《现代计算机》2020年第7期18-23,28,共7页Modern Computer

摘　　要：随着社会和经济的发展,信用危机越来越受到重视,它不仅影响个人的信用还对很多金融机构产生一定的影响。在信用风险模型中,信用违约率是一个主要的方法来评估个人的信用。机器学习方法一直被用来评估个人违约率。然而,在计算个人违约率的过程中,个人信用数据的缺失是我们面临的一个主要问题。通常,为了更精确地计算个人违约率,对缺失值进行填补是一个比较实际灵活的方法。引用一种特征权重灰色K近邻(FWGKNN)新型算法来对缺失数据进行填补进而求得更精确的个人违约率。实验结果证明:相比其他三种算法,基于FWGKNN算法求得的个人违约率更加精确。With the development of society and economic, credit risk has risen popularly, affected both institutions and individual. Probability of default(PD) is one of the major measurements in credit risk modelling used to evaluate personal credit. Machine learning approaches can be easily used for the consistent estimation of individual PD. However, the problem of missing data is always ignored when calculating the PD.In general, missing data imputation is an actual and feasible way to calculate the PD accurately. In this paper, we introduce a new method named feature weighted grey K nearest neighbors(FWGKNN) algorithm to impute the missing data in order to calculate the PD accurately.An experiment is conducted to verify the performance of the method and three approaches are used to compare the imputation results and the prediction accuracy about PD. Experimental results show that our method is considered superior to the other three comparative approaches, and the calculated PD based on our method is more accurately.

关 键 词：信用风险 违约率 数据缺失 特征权重灰色K近邻算法 

分 类 号：F832.4[经济管理—金融学] F224