The Application Of Logistic Regression And Related Methods In Personal Credit Scoring

In recent years,with the rapid development of economy,people's consumption concept has also changed.People tend to more and more consumption in advance.More and more people apply for loans from financ ial institutions such as banks or commercial companies.And the applicant often pay attention to whether was granted loans or not,at the same time,the bank and other financ ial institutions are focused on whether the applicant will be effected according to the predetermined time to repay the loan.They evaluate the applicant's personal credit score by using credit scoring model.In this way,they can judge whether the loans granted to the applicants,and also predict the applicant whether a default or credit customer.The credit policy makers have been concerned about how to minimize the loss of banks and other financ ial institutions and obtain maximum profits,therefore,it is particularly important to establish a suitable and effective credit scoring model.The imbalance personal credit data sets was used in this paper.In the non-equilibrium data preprocessing,the random over-sampling in resample was adopted.The model was applied to select factors that affect personal credit scoring,the traditional Logistic regression method,and the improving Lasso-Logistic regression on the Logistic regression method,the Adaptive Lasso-Logistic regression three methods,Lasso-Logistic regression add penalty term to the likelihood function,the Adaptive Lasso-Logistic regression apply weight on penalty terms,and give different punishment on regression coefficients,that is to use adaptive weights to punish different coeffic ients,when calculating the coefficient weight,and choose the Adaptive Lasso in Logistic regression method,the maximum likelihood estimation and ridge estimation,respectively,as the initial estimates,through the analysis of personal credit scoring data,taking the prediction accuracy and minimum misclassification error as a measure standard,using ROC curve to validate the results,analyzing and comparing these methods in the credit scoring of predicted results.Through application of the above several methods on credit data,the practice results show that the Logistic regression and the improved method all have good robustness and interpretability,comparatively speaking,the accuracy of logistic regression is the lowest,misclassification error is also the highest,Lasso Logistic regression penalized the estimation of Logistic regression,it chose the relatively small number of variables,reduce the complexity of the model,and improve the prediction precision of the model,reducing the misclassification error.The Adaptive Lasso-Logistic regression method give adaptive weights for each coefficient,and the initial estimated selects the maximum likelihood estimation,the model shows the best prediction accuracy and the minimum misclassification error,and the same time,the type I error and the type II error are reduced to the minimum.When selecting the ridge regression estimation as the initial estimation,the model also has a good performance,second only to the maximum likelihood estimate.Whether the classification accuracy or the prediction accuracy,are higher than the Lasso-Logistic regression model,at the same time,the type I error and the type II error,the minimum misclassification error is lower the Lasso-Logistic regression.