Customer Churn Prediction Using Data Mining Techniques for an Iranian Payment Application

Abstract

Customer Relationship Management (CRM) and data-driven marketing have become of paramount importance in this age of evolved markets and fierce competition among businesses. One of the most important branches of CRM is retaining existing customers. Since customer acquisition is about 5 to 6 times more costly than retaining customers, achieving an accurate model for customer churn prediction is essential to devise marketing retention strategies. Therefore, in this study, ensemble models are proposed to predict customer churn. Since customer churn is a rare occurrence in an organization and causes an imbalanced distribution in the target variable, ensemble learning algorithms, one of the most efficient and widely used methods, have been used to deal with this problem. With regard to the case study, the dataset was generated on demographic and 13-month transactions of users of an Iranian payment application. In this study, the best model to predict customer churn is the bagging version of Decision Tree, reaching the highest accuracy, f-measure and AUC.