Sampling Method for Imbalanced Distribution in Customer Churn Model

In broadband customer churn early-warning model, the training samples are usually imbalanced, which leads to poor performance of the model built. And the two traditional sampling methods (over sampling and under-sampling) both have their benefits and drawbacks. Therefore, this paper proposes a new improved sampling method, utilizing both the advantages of the traditional sampling methods. And the experiments demonstrate that the new sampling method improves the accuracy and efficiency of customer churn early-warning model to a certain degree.