Research On Credit Model And Parameter Update Based On Logistic Regression

Financial inclusion is an important strategic decision in China during the 13 th FiveYear Plan.With the rapid development of the Internet,the rate of bad debts due to personal credit issues is increasing year by year,and how to accurately assess personal credit has become particularly important.This paper actively responds to the call of national and regional strategies,relying on the "three rural" digital universal financial services platform of Gansu Bank,based on the "three rural" big data of Gansu Province,using big data technology,the study of personal credit problems of Gansu Province farmers.First,the data pre-processing work,including missing values and outliers,was carried out on the basic data of farmers and historical business data provided by Gansu Bank,and then the Pearson coefficient,Spearman coefficient,principal component analysis(PCA)and other methods were applied in a comprehensive manner.Second,based on the characteristics of the selected experimental samples,the credit model was constructed using Logistic regression,decision tree,K proximity,random forest,and support vector machine(SVM)methods,respectively,and the prediction results were compared based on the evaluation criteria of accuracy,AUC,time spent,etc.The comparison found that the Logistic regression algorithm had the best combined effect.Therefore,we use a Logistic regression algorithm to predict the amount of credit granted to farmers and calculate the final amount of credit granted to farmers based on the pregranted credit limit,combined with the credit report provided by the People's Bank of China.Finally,because the model training is directly applied to the actual production after the completion of the existing research,the parameters are no longer updated continuously,which leads to a decrease in the accuracy of the model prediction and an increase in the bad debt rate.The experimental results showed that the accuracy rate of logistic regression reached 81.695% by default,while the accuracy rate of the model reached 96.486% and AUC reached 96.946% after the model parameters were updated,which obtained better results and achieved the purpose of maintaining high prediction accuracy of the model.At the same time,through the realization of the model system,the work mode of the whole process of the farmers' loan online completion.