Post-loan Risk Rating Of The Internet Micro Lending Based On XGBoost

With the change of the national consumption concept and the progress of internet financial technology,the internet micro lending industry has achieved rapid development.There has been a significant increase in the number of companies providing internet micro lending,the size of customers,and the amount of loans.In order to ensure the healthy,stable and sustained development of the industry,it is the key for the internet micro lending company to raise the risk management level of credit loan.Based on the above background,this paper takes the loan data obtained from Lending Club as the research object,establishes a post-loan risk rating model based on the XGBoost algorithm from the perspective of practice,and studies the risk factors that affect the probability of customer default.Firstly,based on business knowledge,this paper analyzes the development history and current status of domestic internet micro lending,including changes in micro lending methods and domestic regulatory policies for the internet micro lending.Through analysis,this paper points out the importance of establishing a reasonable and efficient credit risk assessment model.Secondly,this paper builds a post-loan risk rating model based on the XGBoost algorithm in machine learning.The second-order expansion of Taylor function is used to approximate the loss function to increase the speed of model training;L2 regularization is used to reduce model complexity;Greedy algorithm is used to measure the splitting conditions and the post-pruning of trees.The sparse sensing algorithm is used to automatically learn the splitting direction of missing data,which effectively maintains the distribution characteristics of the data;A distributed weighted histogram algorithm is used to find the split nodes.Then,aiming at the development process of the post-lending risk model,this paper takes the loan data obtained from LendingClub,an overseas lending platform,as the research object.According to the business knowledge,the exploratory analysis is performed on the characteristic variables in the data,and the missing values,abnormal values and unbalanced samples in the data are processed.Based on the analysis of feature relevance and the importance of features,the variables used in model training are screened.Finally,the control variable method is used to determine the optimal solution of the model training parameters,and these parameters are used to train the model.The training results show that the constructed post-lending risk rating model can obtain high AUC and F1 values,which validates the effectiveness of the model.Based on the experimental results,this paper analyzes several important factors that affect the probability of customer default,which provides reference for practical applications.