Research On The Overdue Risk Rrediction Of Credit Card Of A Rural Commercial Bank Based On Catboost Model

Since China’s banking industry began to develop credit card business in 1985,the scale of credit card business has been expanding.Rural commercial banks,as an indispensable main body in the development of China’s financial market,also actively enter the credit card market.However,when rural commercial banks enter the credit card market,they have to face the market competition of major commercial banks and the challenge of the ability to control the overdue risk of credit cards.In the management of overdue risk of credit card,the agricultural commercial bank mainly uses the form of expert score card to score the overdue risk of customers,but this method has a strong lag and ignores the characteristics of customers of agricultural commercial bank,so it can not predict the overdue risk of credit card customers well in practical application.At present,in the Internet financial enterprises,data mining methods are widely used,and take a certain effect.There are many data mining methods,such as support vector machine,random forest and so on.Although good results have been achieved in some scenarios,it is difficult to further improve the prediction accuracy.At present,in the field of machine learning,integrated learning method is a research hotspot.The integrated learning method can improve the performance of weak learners,among which catboost model is a relatively new integrated learning model,which has a further improvement in prediction performance and has been applied in various fields.Therefore,this paper uses catboost model to predict the overdue risk of rural commercial bank’s credit card,which provides some reference for improving the accuracy of the overdue risk prediction and the risk control level of rural commercial bank’s credit card.This paper studies the overdue risk prediction of rural commercial bank a credit card.Firstly,this paper studies the theory of information asymmetry,bounded rationality and behavioral economics.Based on the theoretical research,it analyzes the factors that affect the overdue behavior of credit card,including the factors of consumption behavior,product,attribute and information.At the same time,it establishes the index system to describe each factor.Secondly,this paper analyzes the shortcomings of current credit card overdue risk assessment methods of rural commercial bank,and constructs a catboost model based on particle swarm optimization to predict the overdue risk of credit card.Third,this paper collects the credit card transaction data of ABC a,makes an empirical analysis of catboost model,and compares it with SVM,random forest and BP neural network.The empirical results verify that catboost model has improved in various performance indicators.Finally,through the calculation of catboost model on the importance of indicators,this paper analyzes the indicators that have an important impact on the prediction and makes a theoretical analysis.The research results of this paper are as follows:(1)in the current credit card overdue risk assessment methods of rural commercial bank,the expert rule scoring method is mainly based on the experience of experts,the data indicators used are limited,and has a large lag,unable to reflect the dynamic behavior and state changes of customers.From the perspective of the actual prediction effect,the overdue rate of credit cards of various rural commercial banks is relatively high,which shows that this method is difficult to truly reflect the overdue risk of customers.However,at present,the data mining model studied by the academic community is not stable,and the prediction based on a single model is easy to encounter performance bottlenecks,so it is difficult to make further optimization;(2)from the empirical analysis results,the catboost model constructed in this paper has the best prediction effect outside the sample,compared with other models in various performance indicators have been improved;(3)from the impact of overdue risk prediction From the perspective of the importance of the indicators,the scores and rankings of credit investigation and data integrity of PBC are higher.From the perspective of information asymmetry,data collection of customers in the financial system is an important factor in predicting overdue behavior.From the perspective of consumption behavior,the amount and number of transactions of the past 7 days and 30 days also have an important impact on the prediction of overdue behavior.Abnormal consumption in a short period of time may indicate that the cardholder has excessive consumption based on irrational behavior.From the perspective of "psychological account" and "self-control",the cardholder has irrational decision-making and eventually overdue behavior.The interest rate,repayment date and installment of credit card also have some influence on the overdue behavior,which shows that "naive delay" and "mature ahead of time" also lead to the occurrence of overdue risk of credit card to some extent.The study of this paper has a certain reference for further improving the prediction level of credit card overdue risk of rural commercial bank a,and also provides a reference for other rural commercial banks to manage credit card overdue risk.