A Research Of Personal Credit Rating Evaluation Based On Genetic XGBoost Model

With the rapid development of mobile internet,all walks of life merging with the concept of "Internet +" generate the new creativity.the P2 P online loan,one of the mainstream development models of Internet finance,relies on the efficiency of Internet finance has entered a stage of rapid development.However,the P2 P industry is a new industry,the threshold of employment is low,the quality of employment is uneven,and the credit risk arising from the inability to effectively assess the credit level of borrowers due to information asymmetry is rapidly gathering on the platform.The P2 P industry will inevitably produce an adverse selection if the industry still cannot effectively assess the default risk of borrowers.Based on this background,this paper takes online loan default users as the research object,and uses the datasets provided by UnionPay Big Data Business Platform on various aspects of borrowers to conduct research on personal Internet loan credit evaluation through modern statistical techniques and data mining techniques.The research in this paper is mainly divided into four parts: First,the literature is used to define the concept of Internet finance,which leads to the P2P(Peer to Peer)online loan model study.Secondly,in order to select the credit evaluation variable Scientifically,and have a solid theoretical basis,I review the relevant literature on credit evaluation,select and improve on the basis of previous studies,and select effective credit evaluation variables applicable to Internet finance.Third,through combing the XGBoost model theory,construct a genetic XGBoost credit evaluation model based on The XGBoost algorithm combing with genetic algorithm.Fourthly,it compares with the advantages and disadvantages of traditional models to explore various application scenarios.Finally,it analyzes the economic behavior characteristics of credit default customers according to the importance of the characteristics of the model.The research results show that the default risk of the borrower is significantly different from those factors,which are other lenders' last lending amount to the borrower,the standard deviation coefficient of the borrower's transaction amount in the past six months,the average amount of the borrower's transaction in the near-term in June,and the age of the user;through the comparison of various traditional classification models under the same dataset conditions,the classification performance of genetic XGBoost model is better than logistic regression and Gaussian kernel SVM model,which is suitable as a personal credit loan evaluation model with category imbalance.However,the SVM model is not suitable for Internet financial credit risk assessment characterized by large data volume and high latitude due to large calculations and the inability to quantify the probability of default.