Research On User Portrait Modeling And Ensemble Algorithm In Personal Credit Field

With the continuous development of China's economy,the domestic financial credit industry has developed rapidly,and many financial credit services have gradually been promoted to the public life.Personal credit business has already flourished.With the continuous development of Internet finance,personal credit consumption is increasing.In general,the establishment of user portrait models and overdue recognition models in the credit industry has become an urgent need.This paper first puts forward the idea of variable subdivision for the basic sample data,and obtains eleven dimensions of variable subdivision and based on the traditional RFM value evaluation model,we established a comprehensive evaluation index system include the three dimensions of credit user spending power,user stickiness and repayment willingness.Based on the variable subdivision dimension and the comprehensive evaluation index system,we have established a user portrait model,which includes user segmentation portraits and user funnel portraits.For the user segmentation portrait,we use the comprehensive evaluation index system to subdivide the user groups in the credit field in the three dimensions of consumption ability,user stickiness and repayment willingness,and evaluate the value of each type of user group and propose correspondingly.For the user funnel image,we use the user funnel algorithm based on the eleven dimensions obtained after the variable subdivision dimension,and get the high overdue ratio user group and capture the behavior path of the high overdue ratio user group.After building the user portrait model,we focused on building an overdue recognition integrated algorithm model.Firstly,based on the idea of variable subdivision and related machine learning algorithms,the scores and dimensionality reduction of the basic sample data are completed,and two new types of sample data are obtained,Full-featured scoring data and subdivision dimension scoring data,and studied the influence of different sample data on the prediction accuracy of random forest integration algorithm.Based on two different samples and five different integration algorithms,the optimal Stacking integration algorithm is selected by evaluating test sets,cross-validation sets,learning curves and several evaluation indicators,and provides a accurate,stable,and scientific overdue risk identification models.This paper constructs the user portrait model in order to better serve the business.The integrated algorithm model is built to better serve the overdue risk identification accuracy.