Research And Application Of Internet Financial User Churn Prediction Model Based On Data Mining

With the rapid development of network technology,there are more and more Internet financial platforms,and network financial products have become seriously homogenized.The possibility of user loss is increasing,and user loss will have a great impact on enterprises.Relevant research data show that the cost of acquiring new users is eight times that of retaining old users.Moreover,the company's annual maintenance rate for users can increase by 5% and the company's profits can increase by 85%,so reducing the loss of users is very important for the company.The research focus of this thesis is to use Date Mining(DM)technology to analyze the data of Internet financial users and predict the loss of users through classification model.The main work and achievements of this paper are as follows:1.The thesis first elaborated the current research status,existing problems and application status of data mining technology in the management of user churn in the financial industry.The current mainstream data mining technology related algorithms and modeling process.2.Based on the data of lost users in Internet finance,21 variables consisting of user attributes,asset factors,and behavioral factors are selected.Through the pretreatment of user data,descriptive statistics,chi-square test and Spearman correlation analysis,the characteristics of lost users are studied.At the same time,the impact of various variables on user churn was tested,and finally 13 variables that have an important impact on user churn were obtained.3.The unbalanced data sets are processed by under-sampling method,over-sampling method and SMOTE method respectively.And the prediction effects of three methods are compared in the decision tree model,the logistic regression model,the SVM model under the four kernel functions,and the BP neural network model.The experimental results show that the three sampling methods based on SMOTE method have the highest accuracy,and the BP neural network model has higher accuracy in the four models,so it is more suitable for the Internet financial user churn model.4.Through the ROC curve evaluation of the model and the confusion matrix analysis,the accuracy of the model was evaluated.Apply the model to reality and observe the churn prediction effect of the churn model on users.Finally,use the company's precision marketing platform to carry out targeted loss recovery marketing,to minimize the loss of company users,improve the loyalty of the overall platform users,and enhance the competitiveness of the company.