Research on Prediction Model of Financial User Churn Based on Data Mining

Today, with the rapid development of network technology, a large number of online financial services have shown a high degree of homogenization trend, and a large number of customers will face a large number of losses, which will also cause a great impact on the company. According to relevant investigations, the cost of attracting new customers is eight times more than that of retaining old customers. In addition, if a company's customer reten-tion rate can increase by 5%, the company's profit will increase by 85%, so it is very criti-cal for the company to reduce customer losses. This paper mainly uses data mining tech-nology to make statistics of network financial data, and uses classification mode to predict it, so as to formulate retention strategies for enterprises and reduce the loss of users. Un-balanced data sets are processed by under-sampling, over-sampling and SMOTE methods respectively, and the prediction effects of three methods in decision tree model, Logistic re-gression model, SVM model under four kernel functions and BP neural network model are compared. The experimental results show that the model based on SMOTE method has the highest accuracy among the three sampling methods, and the BP neural network model has higher accuracy among the four models, so it is more suitable for the Internet financial user churn model.