Research And Application Of Overdue Repayment Prediction For Bank Customers

With the rapid development of economic globalization and information technology,the data generated by financial activities are growing rapidly.How to make use of these data through data mining has become the focus of people's attention.At the same time,people's consumption concept is also changing.The increase of consumer credit has not only brought profits to financial institutions,but also created risks.How to control credit risk effectively and become the main target of financial institutions.Due to the characteristics of the credit industry data,the data set is prone to imbalance.When dealing with imbalanced data sets,the performance of traditional classifiers will begin to decline in different degrees.How to improve the method of dealing with imbalanced data sets and improve the efficiency and accuracy of bank credit approval process by quantifying customer information has become one of the main concerns of Banks.In this paper,based on improving random forest algorithm(IRF)is used to study the problems related to the prediction of overdue repayment of bank customers.This paper first introduces the development of overdue repayment prediction of bank customers,then introduces the theoretical basis of stochastic forest model and the method of unbalanced data set processing.Secondly,the problems in the treatment of unbalanced data sets of the random forest algorithm were improved.By comparing the Breast Tissue and Glass data sets in UCI,the differences and merits of the improved random forest algorithm,support vector machine,classical random forest,traditional logistic regression and weighted random forest in the treatment of unbalanced data sets were compared.Finally,the improved random forest algorithm is applied to empirical studies of bank customer payments are forecasting problem,filter variables and data preprocessing on kaggle website for customer payments,according to the practical work experience to add new variables,set up payment forecast model,then optimize the IRF model,results analysis and compared with the rest of the various models,also verify the IRF performance.The empirical results of this paper show that IRF algorithm has certain effectiveness and feasibility in the processing of unbalanced data set classification problems,compared with support vector machines,classic random forest,traditional logistic regression,weighted random forest that IRF algorithm has certain advantages in the goodness and stability of classification performance in different data sets.At the same time,through combining with the research and analysis of the bank customer overdue repayment prediction model,it shows the value of the algorithm in practical problems.