Enhanced Subscriber Churn Prediction Model for the Mobile Telecommunication Industry By

Abstract

Subscriber churn is a major cause of worry of many industries which require low or zero switching cost. Telecommunication industry can be considered as the most affected and top the list with approximate annual churn rate of 30%. Recently Mobile Network Operators (MNOs) have implemented customer relation management with intention to reduce the number of Subscriber churn, but it is still faced with high churn rate in the industry. It is important to recognize the potential churners before they churn. At this era of Big Data, the telecos have the advantage of using user generated data to predict customer churn. Service usage metrics such as account ID, service ID, Activation date, Deactivation date and others like network performance indicators and traditional demographic information such as Zip code, Age, Sex, population density, cell site coverage are employed by MNOs for churn prediction. The challenge lies in developing effective prediction techniques, this work is aimed at using the Genetic Algorithm for optimal selection of churn attributes from call detail records (CDR) and Artificial Neural Network for churn prediction based on the selected attributes. The WEKA (Waikaito Environment for Knowledge Analysis) tool was used for this work.