Telecommunication subscribers' churn prediction model using machine learning

• Published in: Eighth International Conference on Digital Information Management (ICDIM 2013) pp. 131 - 136
• Main Authors: Qureshi, Saad Ahmed, Rehman, Ammar Saleem, Qamar, Ali Mustafa, Kamal, Aatif, Rehman, Ahsan
• Format: Conference Proceeding
• Language: English
• Published: IEEE 01-09-2013
• Subjects: Business Intelligence; Churn prediction; Correlation; Data Mining; Decision trees; Linear regression; Logistics; Prediction algorithms; Predictive models
• DOI: 10.1109/ICDIM.2013.6693977

During the last two decades, we have seen mobile communication becoming the dominant medium of communication. In numerous countries, especially the developed ones, the market is saturated to the extent that each new customer must be won over from the competitors. At the same time, public policies and standardization of mobile communication now allow customers to easily switch over from one carrier to another, resulting in a fluid market. Since the cost of winning a new customer is far greater than the cost of retaining an existing one, mobile carriers have now shifted their focus from customer acquisition to customer retention. As a result, churn prediction has emerged as the most crucial Business Intelligence (BI) application that aims at identifying customers who are about to transfer their business to a competitor i.e. to churn. This paper aims to present commonly used data mining techniques for the identification of customers who are about to churn. Based on historical data, these methods try to find patterns which can identify possible churners. Some of the well-known algorithms used during this research are Regression analysis, Decision Trees and Artificial Neural Networks (ANNs). The data set used in this study was obtained from Customer DNA website. It contains traffic data of 106,000 customers and their usage behavior for 3 months. We also discuss the use of re-sampling method in order to solve the problem of class imbalance. Our results show that in case of the data set used, decision trees is the most accurate classifier algorithm while identifying potential churners.