Research on Customer Churn Intelligent Prediction Model based on Borderline-SMOTE and Random Forest

Abstract

In the era of big data, with the rapid development of Internet finance and the increasing pressure on banks to compete, banks have turned their goal to retaining old customers, so it is essential to predict customer churn. Around this problem, this paper designs a Borderline-SMOTE-random forest prediction model for unbalanced data such as bank customers. Combined with the oversampling algorithm, it can better solve the unbalanced data and the strong anti-noise ability of random forest. OOB error rate, AUC, Precision, Recall, and F-mean are used as the evaluation indicators of the model, and KNN, decision tree and Naive Bayes are used as comparisons. The experimental results show that the Borderline-SMOTE-random forest prediction model has the best ability to solve the problem of bank customer churn prediction among the models, and its performance is improved by about 4% compared with other models.