Design And Implementation Of Credit Scoring System

With the rapid and stable development of China's economy,the development of Internet financial credit business has reached a climax.The explosive growth of customer data,the lack of reliability,accuracy and the slow efficiency of mass data processing,the inability to effectively excavate the value of data,etc,are becoming more and more prominent.How to better mine user credit data,shopping data containing information to reduce the occurrence of bad debts,while accurate classification of customers to achieve better Internet financial wind control has become an important research direction.Because of the particularity of the financial industry,in order to lower the threshold of developing the scoring card model and improve the modeling efficiency,there should be a credit scoring system to complete the construction of the credit scoring model.In view of the above problems and needs,through query ingenuity of credit scoring system through query and reading related literature,improve the problems such as THE non-monotony of WOE and the excessive proportion of samples in a sub-box after the card-square division box;The method of gradual regression and other methods to select the module variable solves the problem of the Internet high-dimensional characteristics difficult to select,the Spark technical framework is studied in depth,the realization of a credit scoring system,which is composed of resource management module,model building module,visualization module three modules.The resource management module is composed of data resource management,model management and task process management.It is mainly responsible for data upload and download,model storage and deletion,task search and deletion,model building blocks are composed of functional components and algorithm components.The functional components encapsulate the data processing logic,the algorithm components encapsulate the logic regression algorithm,its main function is responsible for the model fitting before the feature processing and the model training after;Data visualization is responsible for the presentation of intermediate data and model evaluation results,and process visualization is responsible for the presentation of the entire modeling process.In this paper,using the credit credit data of a domestic Internet consumer finance company to build a credit scoring model,using the improved card square division box method for feature division and then fitting using the logic regression algorithm,both AUC and KS indicators were improved,in which AUC from 0.7267 to 0.7373,KS rose to 0.353 from 0.339.The credit scoring model is built and run in the credit scoring system by means of components,which reduces the threshold of building credit scoring model and improves the efficiency of building the credit scoring model.