A Model for Customer Churn Management of an Internet Service Provider

Abstract

Customer churning is one of the most important issues facing Internet Service Providers in a competitive and rapidly saturating market. Due to the high costs associated with attracting new customers, ISPs have turned to a customer retention approach that explicitly seeks to reduce churn. This study has been surveyed the churning of internet service customers in one of the largest telecommunications companies in Iran. In order to predict the churn, customer data has been collected during six months, and their churning behavior has been investigated over a period of one year after. In addition to churn prediction, the most important factors affecting churn have been identified. In the preprocessing step, the "Random Under-Sampling" method is used to balance the data set and the "minimum-Redundancy, Maximum-Relevance" method is used to feature selection. Then, the "Random Forest", "Support Vector Machine" and "K-Nearest Neighbors" algorithms were applied to classify churning and non-churning customers, and the evaluation criteria show the superiority of the random forest algorithm. The final model, which was obtained from a combination of balancing, feature selection and classification methods, called the RUS-mRMR-RF model, is considered as an efficient model in customer churn prediction and identifying the most important factors affecting churn. The results of this study provide valuable insights to the company to develop customer retention strategies.

Keywords

• customer churn prediction
• feature importance
• random forest
• internet service
• telecommunication company