摘要
针对信用风险评级过程中数据的高维性和各属性数据存在量纲差异性的问题,提出了一种基于邻域粗糙集和距离判别的信用风险评级方法。邻域粗糙集是一种常用的降维方法,与传统的降维方法相比有严格的数学推导,对数值型属性和类别型属性可作分别处理。数值型属性在距离判别时容易受到数据量纲的影响,采用马氏距离可以消除数据量纲差异对分类造成的影响。因此以待测样本与各类训练总体之间的马氏距离的大小作为判据对待测样本进行分类。对现实数据进行实验,各类和总体的数据准确率表明该方法是一种有效的评级方法。
In essence, credit risk evaluation is a credit risk evaluation are always high dimensional valuation method based on distance discrimination to credit risk evaluation. Neighborhood rough set problem of pattern recognition. The data applied in and quite different in dimensions. We present an e- and neighborhood rough set and applied this method is a common method of dimension reduction and it has strict proof. It is valid to both numerical attributes and categorical attributes. The difference of the dimensions is important to the accuracy of classification, we adopt Mahalanobis distance which liminate the influence of the dimensions. In the paper we regard the Mahalanobis distance betw tested sample and the training set of every class as the criterion for classification. Through the a real data set, the accuracy rates of every class and total tell us that that this method is an e method for classification can eeen the test on ffective
出处
《重庆理工大学学报（自然科学）》
CAS
2013年第2期130-134,共5页
Journal of Chongqing University of Technology：Natural Science
关键词
信用评级
信用风险
邻域粗糙集
马氏距离
credit evaluation
credit risk
neighborhood rough set
Mahalanobis distance
