Research Of Individual Credit Model On Adaptive Penalized Logistic Regression

Since the 1980s in China,the credit card business has made great progress.With the rise of Internet credit loan,the traditional credit loan has experienced a new round of growth,which aroused the attention of customer credit problems.The importance of data mining in the credit card business is increasingly apparent.1.E-Wilcoxon penalized Logistic regression algorithm was proposed by improving Wilcoxon penalized Logistic regression algorithm.Based on Wilcoxon rank sum test,the rank can be used to assess the ability of different characteristic to distinguish different categories.The penalty formula of adaptive weight was changed by exponentiation,so the punishment effectiveness was strengthened.And the convergence of E-Wilcoxon penalized Logistic regression algorithm was analyzed.2.The empirical background was based on the credit risk assessment of credit card customers.The algorithms models of adaptive penalized weights were calculated by using Python,and the feature variables were analyzed.Six evaluation indexes were selected to assess the models,which included accuracy.precision,recall,F1-value,G-mean and area under the ROC curve called AUC score.Then the experimental results of these models were comparatively analyzed,which consisted of Logistic regression,Wilcoxon penalized Logistic regression and E-Wilcoxon penalized Logistic regression.The comprehensive evaluation showed that the Logistic regression algorithm was not as valid as the penalized Logistic regression algorithm.The E-Wilcoxon penalized Logistic regression algorithm proposed in this paper was higher than the Wilcoxon penalized Logistic regression algorithm,which could be found from all values of the above six indexes.And the values were respectively increased by 6.17%,7.58%,2.81%,6.36%,5.19%and 7.47%.It was demonstrated that the improved algorithm was superior to the algorithm of the original one.