Loan Forecasting Based on Machine Learning and Variable Correlation Analysis

With the economic development and social progress, loan has become one of the most common business of all kinds of modern banks. In order to reduce the risk of loan, it is particularly important to put forward an effective method to forecast loan. Therefore, this paper proposes a loan forecasting method based on Xgboost and feature importance evaluation. Specifically, after cleaning the original data set, we evaluate the classification effect of several different machine learning algorithm models and find that the Xgboost effect performs best. In addition, in order to enlighten the bank staff, in addition to predicting the results, we also use SHAP and feature importance method to carry out importance analysis on the relevant feature variables, and explain the influence of each feature on the model output. The results show that the principal paid by the borrower, the amount of financing and the amount of recovery after the borrower defaults are the main factors affecting the default risk. The results and methods of this study can provide important references for financial institutions in the process of loan approval, and help to improve the effectiveness and accuracy of risk management.