Identification And Prediction Of Default Risk Factors Of Personal Auto Loans In Auto Finance Companies

Automobile finance companies are the main platform for providing automotive consumer credit,and personal auto loans are the main way for users to obtain car loans.Consumer automotive credit is an important part of automobile finance.Traditional car loan approval processes involve manual review based on user-provided personal information,which is time-consuming and laborious.Over the past few years,the increasing income of Chinese residents has led to a significant rise in demand for consumer automotive credit,resulting in a continuous expansion of the scale of automotive credit loans.The traditional car loan approval model is not suitable for the current market environment,and establishing a suitable car loan default prediction model can help expedite loan approval.The recent alterations in the automotive credit market environment owing to the impact of the COVID-19 pandemic have led to escalated non-performing loans in automobile finance companies as opposed to commercial banks.Therefore,developing a reasonable car loan user default identification model is an urgent problem to be solved.This study analyzes customer data from automobile finance companies,develops a prediction model for car loan default among users,the study is organized into six sections.The first section of the paper provides an introduction to the research background and significance.The second section presents a summary of relevant theories and achievements in the field of automotive finance and car loan default prediction.The research content,methods,and framework of this paper are also described.Thirdly,automotive finance-related theories,machine learning-related algorithms,and methods for handling imbalanced data are introduced.Fourthly,three single learning algorithms and three ensemble learning algorithms are used to predict whether car loan users will default,and grid parameter optimization is performed to select a suitable classification algorithm.Fifthly,the ensemble classification algorithm is improved using methods for handling imbalanced data,and the improved model is used for empirical comparative analysis.The final section of the paper presents a summary of the research results and proposes reasonable suggestions.The results of the study show that the overall performance of ensemble learning algorithms is better than that of single classification algorithms.However,the difficulty of parameter tuning is higher,and the fitting time required is longer.The Borderline method is more effective in handling imbalanced data than the SMOTE and random oversampling methods.BSO2 performs the best in the random forest model,while BSO1 performs better in the XGboost model and Light GBM model.The SMO method also has a significant effect on improving the classification performance of the model,but overall is not as good as the BSO method.The RO method has the smallest effect on improving the classification performance of the model.In the improved model,the BOS1-Light GBM model demonstrates superior performance in identifying default risk,achieving an accuracy of96.72% and a recall of 87.24%,has the best overall performance in identifying default risk,with an accuracy of 96.72%,a recall rate of 87.24%,a specificity of 97.58%,an F1 value of81.59%,and a G-mean of 92.27%.Among all the features,asset cost,loan-to-asset ratio,credit rating,past six months default quantity,and total monthly payment amount have the highest contribution rates.Drawing on the findings of this study,the following suggestions are put forward: 1.Enhance the automobile finance risk management framework and foster sustainable growth of the automobile finance sector;2.Improve the user credit review mechanism and integrate personal information across platforms;3.Strengthen the management of the automobile finance industry and improve relevant laws and regulations;4.Implement domestic automobile finance policies and determine development directions.