Research On Forecasting Accuracy Of Bank Credit Card Customer Default Probability Based On Data Mining Technology

Credit card first appeared in Europe and the United States and other developed countries in the 1960 s,and today it have become one of the most important payment tools.Since China's reform and opening up,credit cards have gradually ushered in a consumer boom and rapid development.However,due to a series of uncertainties such as default by credit card customers and loopholes in the banking system,accounts receivable cannot be recovered in time,that is,some or all of the defaults or bad debts may occur.This has gradually become a huge hidden danger of major commercial banks and financial institutions..In order to solve the above problems,this paper proposes a research on the accuracy of bank credit card customer default probability prediction based on data mining technology.In fact,credit card customer default assessment can not only help commercial banks to reduce or even eliminate the capital risk of credit business,strengthen the maintenance of customer information and credit business management,but also conducive to the overall prosperity and stability of China's financial market.In this paper,firstly,the Bat Algorithm is optimized by adjusting the dynamic adaptive weight and updating the bat position by Cauchy mutation,A Bat Optimization Algorithm based on dynamic adaptive weight and Cauchy mutation is proposed.The experimental results show that the convergence speed and accuracy of the algorithm are obviously better than the original algorithm and some other improved algorithms.Then,the feature selection method is optimized,and a kind of personal credit risk assessment indexes based on bank customer data is proposed and established.The Bat Optimization Algorithm is used to combine BP neural network.Aiming at the constructed risk assessment index,a new combination model of multiple classifiers for credit card default prediction is trained..Based on the in-depth analysis of the current bank credit card default prediction technology,this paper uses the idea of machine learning to clean and preprocess the data set with the help of the personal credit data of 30000 customers of a commercial bank in Taiwan.Split the data into training sets and test sets.There are 23 independent variables in each customer's information.The feature variables are adjusted according to the correlation and importance of each variable,and a variety of data mining methods are used to model the training set.By comparing the accuracy of these models in customer default prediction,four classifiers with better prediction effect and obvious differences are selected as model combination.Finally,the trained combination model is compared with the base classifier model,and it is found that the Bat Optimization Algorithm based on dynamic adaptive weight and Cauchy mutation combines BP neural network training multi-classifier combination model has a higher prediction accuracy and a lower false alarm rate in the bank default test data,.This paper provides a new way to solve the problem of credit card customer default probability prediction in commercial banks.