Research On Auto Loan Credit Risk Model Based On Multi-source Data Integration Analysis

The advent of the Internet era has given new opportunities for the development of various industries.For the auto finance industry,more and more car loan platforms have emerged.With the development of the auto loan business,due to the uneven credit qualifications of customers,defaults are not uncommon.How to identify the risk of customer default has become particularly important.At present,scholars have conducted research on the credit risk of car loans based on the perspective of the network platform,while few scholars have conducted analysis based on the perspective of the bank.The bank’s car loan business usually cooperates with multiple car sales platforms to issue loans to customers on these platforms,that is,it is necessary to predict the credit risk of customers from different channels.These customer loan purposes are all related to vehicles,and they come from different platforms.The operating models of each platform and the source of customers are different,so there are both similarities and differences.This article treats car loan customer data from different platforms as multi-source data and uses integrated analysis to solve it.At the same time,since car loan credit risk data is usually unbalanced data,the imbalance feature is taken into account in the integrated analysis.Construct a logistic regression model to deal with the two-classification problem,and consider the cost-sensitive learning method based on the composite MCP(Minimax Concave Penalty)penalty function.The above model uses a two-level variable selection method,and the optimization problem is solved by the group coordinate descent method when the model is solved.In the simulation analysis,the composite MCP model(c-cMCP),which considers cost-sensitive learning,and the model that only considers the composite MCP penalty,are compared based on the model evaluation indicators,Accuracy,AUC(Area Under Curve),and Recall.(cMCP),the sub-data set MCP model(MCPs)and the combined data set MCP model(MCP)predict that c-cMCP is better than the other three methods,especially when the imbalance is high.In the empirical analysis,this paper selects two channels of car loan business to obtain the customer’s PBC credit variables and vehicle information,establishes a c-cMCP-based car loan business credit risk model,and calculates the model’s AUC,KS and recall.The comparison and analysis of models based on cMCP,MCPs and MCP show that the c-cMCP model performs better on the empirical datasets.