Customer churn predictive modeling by classification methods

Abstract

The article describes methods of construction of predictive models for classifying customers based on their churn from the company for the example of a mobile operator. There are roles and tasks of customer analytics for understanding the business behavior of customers. The specificity of customer churn for companies associated with a subscription and transactional business model, involving regular customer payments is discussed, and the main reasons for churn are shown. Particular attention is paid to the analysis of forecasting methods based on classification methods. Here we discuss the forecast models based on the decision tree method and the Bayesian network. The decision tree method is basing on the C5.0 algorithm. The Bayesian model is constructed for a Naive and Markov structure. Customer service has become a key factor in the customer churn in all three models. A comparative analysis of the models was conducted based on indicators AUC and Gini. The decision tree model showed the best results. Moreover, the decision tree model shows the reasons why the customer can leave the company and give information for an individual approach to each customer. SPSS Modeler was used as a tool for building models.