Research On Personal Credit Rating Of Small And Micro Loans Based On Machine Learning

With the increasing demand for diversified financing borrowing,new financing methods besides commercial banks and other traditional financial institutions have gradually emerged,among which the development of small and micro online loan platforms represented by P2 P is the most typical.It is undeniable that they uphold the idea of inclusive finance,to a large extent,promote the turnover of capital,alleviate the financing difficulties and expensive financing of small and micro enterprises and individual investors;But it also increases the difficulty for the regulatory authorities to assess the credit of borrowers and control the credit risk of the platform at the same time.In addition,the inherent defects of the platform itself are difficult to solve,which leads to the explosion and withdrawal of online loan platforms on a large scale.In the current environment where data mining and machine learning are widely used,the evaluation of personal credit quality by algorithm improvement and optimization model has become the focus of research.Based on the above analysis,a Stacking algorithm model with Logistic,decision tree,neural network as the base classifier and decision tree as the secondary classifier is constructed.126,090 samples and 16 indicators on the RENREN DAI platform were selected and the optimal indicator system composed of 12 indicators was selected through MRMR to verify the accuracy of the model effectively and compare the performance of the model with that of a single classifier.On this basis,a Logistic credit scoring card model incorporating WOE of evidence weight was further constructed.According to the information provided by customers before the loan,the 12 indicators obtained after the index screening are used to score customers’ credit,so as to help the online loan platform facing the future transformation to predict customers’ maturity default and reduce the credit loss of the platform.The results show that compared with Logistic,Neural Network,Decision Tree and other single classifiers,ensemble learning has better performance in accuracy and stability,and is a more ideal method to predict borrower default status.After a relatively stable and accurate prediction model,Logistic credit score card model can further for the borrower,its overall scored higher on the borrower default rate is relatively low,and sorted by score from high to low,during the same interval of the corresponding borrower default rate trend of increasing faster and faster,thus effectively identifying borrower defaults,It provides a new idea and method for the management and control of credit risk of small and micro loans in China.