Design And Analysis Of Personal Credit Scorecard Based On Logistic Regression

The rapid development of the Internet has changed the traditional industrial structure and promoted the rapid rise of financial business while benefiting all walks of life.The credit business of commercial banks is facing great opportunities,challenges and risk control pressure.Emerge in an endless stream of new technologies under the "Internet plus finance" mode,and new service modes change rapidly,and they are gradually subverting the traditional operation mode of the financial industry.In the era of big data,the judgment of a person’s risk is becoming more and more multidimensional.The traditional financial industry dominated by commercial banks urgently needs to change its business mode,improve the approval efficiency of credit products and do a good job in risk control.Therefore,how to optimize customer classification,provide data support for managers more reasonably and carry out risk management more efficiently has very important practical significance.By querying and reading relevant literature,this paper understands the development process of credit scoring technology at home and abroad.At the same time,a quick survey was conducted on the loan officers of some local banks,covering state-owned banks,joint-stock banks,urban commercial banks,rural commercial banks and rural banks,and learned about the business processes of credit approval of different banks in real life.This paper is mainly discussed through the following five parts: the first part mainly expounds the research background and significance of the topic,research overview at home and abroad,research content and organizational structure.This paper establishes a personal credit scorecard model based on a variety of algorithms in machine learning,so the second part introduces the relevant theories and models in detail.The related theory is mainly the basic theory of scorecard making,and the related model covers four commonly used algorithms in the field of credit scoring.The third and fourth parts mainly focus on the development process of credit scorecard,and discuss it from five aspects: data processing,exploratory analysis,variable selection,model evaluation and scorecard making,and form a comparative experiment.The fifth part summarizes and prospects the work done in this paper.The innovations of this paper are as follows:(1)From the perspective of computer science,combined with the current popular machine learning knowledge,this paper studies the mathematical statistical model,and uses four algorithms such as logical regression,random forest,support vector machine and xgboost to study the algorithm model of personal credit score more comprehensively.(2)In the design process of personal credit scorecard,this paper uses random forest to fill in the missing value,uses smote algorithm to deal with the data imbalance,and uses chi square test in the data box stage to obtain high-quality sample data and give better play to the performance of the algorithm.(3)Through the analysis of four algorithm models such as logistic regression,random forest,support vector machine and xgboost,this paper puts forward a better personal credit scoring scheme,that is,based on the personal credit scorecard based on logistic regression,compound xgboost algorithm and carry out secondary judgment according to the reasonable threshold.It can not only test upward,but also mine potential customers downward.It not only takes into account the stability and interpretability of the logistic regression algorithm,but also makes full use of the accuracy of xgboost algorithm.