A Comparative Study On Credit Default Identification Of Four Kinds Of Data Mining Algorithms

Credit card,as a financial instrument,has made rapid development in the past 30 years.With the rapid development of the credit card market,the number of cards and the credit limit line of the major banks have been improved,and the frequency of the use of the credit card has been greatly increased.Comparing with the European and American countries,China's credit card industry started late.There are many problems in the development process,such as imperfect rules and regulations,low business quality,the unstandardized approval process.Also the extensive operations in order to increase market share caused the risk of credit cards.The existing literature on credit card default mainly concentrated in three aspects.The first aspect is about the causes and prevention of credit risk.The second aspect is about the influence factors and evaluation index of credit card default.The third aspect is about the judgment of credit card default.In order to deal with the financial risks,the Committee on banking supervision,which is involved in G20,has issued a series of financial regulatory measures.The most famous of the regulatory measures is the Basel capital accord ii.In addition,the western information asymmetry theory,cyclical theory and so on have done some explanation to the financial risk.What's more,the existing scoring model,the machine learning algorithm can also help to reduce the risk of credit card to a certain extent.After the reform and opening up,China's credit card business has been developing rapidly,but also brings the risk of default,operation,technology and so on.Among them,the default risk is the key to China's credit card management.It is one of the research topics in this paper to establish a suitable mathematical model to improve the recognition rate of credit card default.First of all,this paper uses simulation data and real data to experiment.Then through the experiment,we compare the risk identification effect of four kinds of common data mining algorithms such as logistic regression,neural network,random forest and support vector machine(SVM).According to the existing algorithms,it is found that the classification of random forest and SVM is better in the comparison experiment of simulated data.Secondly,on the basis of the simulation data experiment and referenced to the kernel function of SVM,we use the independent variables to balance the inner kernel budget and expand the independent variable selection space of the random forest algorithm,which improves the random forest algorithm.This new method optimizes the classification effect of original algorithm.In the real credit card data part,the article introduced the data sources and the experimental environment,and do the cross table between the main independent variables and the dependent variable of the data.Finally,the new method is used to identify the credit card default,and it is proved that the improved algorithm can improve the credit card default rate.