Application Of Random Forests In Personal Credit Evaluation

Recent years,people's consumption concept has changed gradually with national loan policy promulgating one after another and western culture spreading widely.And the scale and scope of personal consumer credit business grows larger and broader.Business like personal housing loans,personal housing renovation loans,car loans,personal daily consumer goods loans,personal travelling loans are all in a dramatic rise.The rise of internet finance even lifts credit consumption to a new level.However,the unsound individual credit system in China and the lagging consumer loans risk management system in credit agencies like commercial banks have caused so many questions and increased credit agencies' credit risk with the development of personal consumer loans business.Against this background,studying how to establish a scientific and reasonable personal credit evaluation index system and how to build a reliable personal credit evaluation model have realistic significance for the healthier development of China's personal loans business.As an excellent machine learning method,random forests has advantages like high prediction precision,ability to process high dimensional data,not easy to overfit and fast training speed,which make it have extensive applications in many fields like text analysis,genomics data analysis,image processing and so on.It can be expected that random forests should be applied into-personal.credit evaluation because of its features.This article takes personal credit evaluation as research object.Firstly,a literature review of personal credit evaluation is conduct and the establishing principle of personal credit index system is summarized based on different opinions on it.Then personal credit evaluation indicators are classified into three groups:personal indicators,economic indicators and credit indicators according to the principle.Secondly,random forests is introduced from several aspects including definition,algorithm,theoretical basis,its advantages and related application,which is theory preparations for the following model building.Thirdly,personal credit evaluation model based on random forests is built after processing and partitioning the German Credit Dataset,and optimized by tuning the parameters:ntree,mtry,and then the paper introduces the ROC curve to evaluate the model.Lastly,the three models:random forests,Logistic and support vector machine are compared in accuracy and stability terms.The results show that random forests are better than the other two models in both accuracy and stability within both German credit dataset and Australian credit dataset,which affirms the applicability of random forests in personal credit evaluation.Meanwhile,feature's importance is evaluated by OOB data in order to make targeted suggestion on banks' loans management job.