Research On Personal Credit Based On Cluster Analysis

With the rapid development of China's socialist economy,especially consumer credit is becoming an important means of stimulating economic growth in China,and people's consumption concept and investment concept have changed dramatically,all kinds of credit business are also increasingly encouraging people to make rational or irrational consumption.With the consumer credit business in today's society in full swing,consumer credit agencies are increasingly concerned about this hot issue of credit ratings,the data of the sample is firstly tested for correlation by independent variables and dependent variables and is classified reasonably,and then we use the related method of credit rating modeling training to explore whether it can be more accurate to predict the credit rating,the article is based on this research.In this paper,we study the independence test of the personal credit rating sample set,and then use the logistic regression and the decision model of the decision tree to make the credit rating forecast after cluster analysis processing to confirm the two guesses:only using the variables that have a significant impact on the credit rating and using all the variables to establish the model to do the prediction of the miscarriage of the difference is very small;whether it can improve the overall accuracy of credit rating after the clustering analysis process of the credit rating data.The main points of this paper are as follows:after the clustering analysis of the credit rating sample is done,the model is used to predict the credit rating "good" and "bad".After the clustering analysis,the credit rating model is used,then the error rate is decomposed into more parameters,that is,using more parameters of the equation to fit the overall error of the sample,it have effectively improved the fitting accuracy.This paper discusses the k-means method in clustering analysis and the method of extending the method based on this algorithm,as well as the common methods of personal credit rating.The former mainly includes k-means algorithm,k-modes algorithm and k-prototypes algorithm.The latter mainly include logistic regression,decision tree and other related methods,and we make further credit rating forecast by the results of the contingency table independent test and the clustering analysis were given in the preliminary analysis,and we obtain the following two conclusions:only the credit rating has a significant impact on the 14 variables and the use of all the variables to establish the model to do the prediction of the overall population and clustering weighted error obtained on the difference is very small;it's certain can improve the overall accuracy of credit rating after the credit rating data doing the clustering analysis and processing.