Analysis Of The Classical Score Card And Machine Learning Models With Southeast Asian Internet Finance Market Data

Since the rise of Internet lending P2P platform in Europe and America in 2005,the development of Internet lending P2P platform in China has experienced a period from high-speed growth to basic saturation in recent ten years,and investors began to seek new markets.With a population of 2.2 billion in Southeast Asia,the Internet industry has achieved leaping development since 2014,and the Internet finance industry is in the high-speed development stage.Compared with the European and American countries with 90%personal credit coverage,the penetration rate of bank accounts in most Southeast Asian countries is less than 40%.There are a large number of people who are not covered by credit and can only seek the help of online credit platform,and the huge market demand of online lending platform appears.In the face of incomplete data and blank credit records of applicants,how to build an efficient risk control model is an urgent problem for risk assessors.There are two ways of pre loan risk assessment in Southeast Asia:manual audit and machine audit.Generally,the manual audit takes more than three days,the risk identification ability is not strong,and the efficiency and accuracy are low.Generally,the way of credit score is adopted in machine audit.The probability of default is predicted according to the historical information of the borrower,then the probability of default is converted into credit score,and finally the decision is made according to the credit score of the borrower.This paper makes an empirical study on the credit risk prediction performance of traditional Scorecard and machine learning methods in Southeast Asian loan market.Based on 44035 loan performance data of real customers in Southeast Asia,this paper first processes a series of data processing steps,such as missing value processing,variable derivation and variable selection,and then uses five algorithms of logical regression,decision tree,random forest,GBDT and XGBoost to make empirical analysis,and forecasts the default probability of application customers and compares them with the actual results.The results show that the machine learning algorithm GBDT has the best performance.Finally,for the traditional logic regression and machine learning algorithm GBDT,the scoring conversion method and scoring results are given respectively.From the empirical results of this paper,in the Southeast Asian market,machine learning algorithm can achieve higher prediction accuracy than the traditional scorecard model,and can also be transformed into credit score as the traditional logistic regression scoring model,which can build a more effective risk control system and provide more effective reference standards for credit reviewers.