Bank Credit Risk Assessment Based On Logistic Regression And GBDT Model

With the development of the modern economy,more and more individuals and companies borrow through banks,and due to some external factors,such as the fluctuation of the financial market,the credit risk is getting higher and higher.At the same time,the development of my country's bank credit evaluation system is not perfect,often relying on the opinions of experts and not intelligent enough.Therefore,for banks,how to establish a credit evaluation system to effectively predict whether the borrower has the ability to repay on schedule is of great significance.Since modern banks often need more mature reliability evaluation algorithms in credit evaluation,this paper proposes a bank credit evaluation algorithm based on a combination of logistic regression and GBDT model.This algorithm uses WOE codes and IV values for feature selection.As the pre-algorithm of the logistic regression model,GBDT algorithm constructs new combination features by the GBDT algorithm,and inputs them to the logistic regression model.Finally,the logistic regression model comprehensively gives the classification prediction results.Aiming at the problem of credit evaluation,the logistic regression algorithm,as a generalized linear model,has good model predictive ability,while the GBDT model can combine the features of the model,and then select the features that contribute more to the model,and at the same time Analyze the importance of the model and give a reasonable explanation for the model's prediction results.This article first introduces the relevant theories of the credit risk assessment model and the model early warning technology,and then through the experimental analysis and verification of the data set published by Ali Tianchi,compared with the single logistic regression model,it has excellent performance and stable effect,and makes the bank's credit risk assessment decision.Provide a quantifiable basis,which can be widely used in the bank's credit evaluation of customers.At the same time,multiple credit evaluation models are used to verify the original data set,and the prediction effects of each credit evaluation model are compared horizontally.Multiple model evaluation standards are used to comprehensively give the evaluation performance of the algorithm.