Hierarchical Alpha-cut Fuzzy C-means, Fuzzy ARTMAP and Cox Regression Model for Customer Churn Prediction

As customers are the main asset of any organization, customer churn management is becoming a major task for organizations to retain their valuable customers. In the previous studies, the applicability and efficiency of hierarchical data mining techniques for churn prediction by combining two or more techniques have been proved to provide better performances than many single techniques over a number of different domain problems. This paper considers a hierarchical model by combining three data mining techniques containing two different fuzzy prediction networks and a regression technique for churn prediction, namely Alpha-cut Fuzzy C-Means (αFCM), Improved Fuzzy ARTMAP and Cox proportional hazards regression model, respectively. In particular, the first component of the hybrid model aims to cluster data in two churner and non-churner groups applying the alpha-cut algorithm and filter out unrepresentative data or outliers. Then, the clustered and representative data as the outputs are used to assign customers to churner and non-churner groups by the second technique. Finally, the correctly classified data are used to create the Cox proportional hazards model. To evaluate the performance of the proposed hierarchical model, the Iranian mobile dataset is considered. The experimental results show that the proposed model outperforms the single Cox regression baseline model in terms of prediction accuracy, Type I and II errors, RMSE and MAD metrics.