Personal Credit Scoring Scheme Planning Based On XGBoost Algorithm

In recent years,with the rapid development of the economy,people's consumption concepts have also changed.People are increasingly inclined to consume in advance.More and more people are applying for loans from financial institutions such as banks or commercial companies.Applicants are often concerned about loan is granted or not.Financial institutions such as banks are concerned about whether applicants can repay loans at a predetermined time.They apply the credit scoring model as a tool for assessing personal credit scores,determine whether to grant loans to applicants,and predict whether applicants are defaulting customers or credit customers.How to minimize the losses of financial institutions such as banks and maximize profits is a problem that credit decision makers have been paying attention to.Therefore,it is particularly important to establish a suitable and effective credit scoring model.This paper focuses on the problems existing in the development of the current personal credit scoring system,attempts to optimize and improve on the basis of the original personal credit scoring system,and accesses the data used in the machine learning competition into the personal credit scoring system,and will select after the indicators are appropriately deleted,the personal credit scoring scheme is eliminated through the machine learning model,thus forming a unified,wellcoordinated and shared personal credit scoring system.The traditional credit scoring is based on logistic regression.The process mainly includes variable binning,WOE(evidence weight)transformation and information Value(IV)of variables,and finally logical regression estimation.Based on the previous discussion of credit scoring and the suggestions of practitioners in the credit field,this paper screens and constructs the index system with the independent variables with large correlation between overdue and overdue,and uses XGBoost to model.Further,with using stacking,it is proved that XGboost has superiority in the fitting effect of the model.Furthermore,for the unbalanced data in this paper,the Ensemble-XGBoost algorithm is proposed to convert the probability into a personal credit score.Finally,the evaluation results of the model are compared with logistic regression,support vector machine,multi-layer perceptron and random forest to verify the rationality and availability of the scheme.