Academic Research on Churn Rate

• Applying data mining to telecom churn management, Hung, S. Y., Yen, D. C., & Wang, H. Y. (2006). Expert Systems with Applications, 31(3), 515-524. Taiwan government removed restrictions on its wireless telecom sector in 1997. The mobile operators started focusing on churn management. This paper evaluates the difference data mining methods to predict the accurate churn rate. The variables it uses are service log, contract, user demographics, call records and billing data. 
• An LTV model and customer segmentation based on customer value: a case study on the wireless telecommunication industry, Hwang, H., Jung, T., & Suh, E. (2004). Expert systems with applications, 26(2), 181-188. Data is collected on the user value for efficient CRM (Customer Relationship Management). This value is estimated on the basis of user LTV (Lifetime Value). However, there are certain limitations, such as the abandonment of customers. This paper follows LTV taking previous profits, abandonment probability and potential advantages into consideration. The customer value helps in identifying the churn rate.
• Fighting churn with rate plan right-sizing: A customer retention strategy for the wireless telecommunications industry, Wong, K. K. K. (2010). The Service Industries Journal, 30(13), 2261-2271.This research offers a customer relations policy for the industry of wireless telecom. The author uses the HFG (Hazard Function Graph) and Pearsons 2 test to calculate the churn rate and the factors affecting it. The empirical analysis shows that the churn rate of the users, who activate optimal package plans, is less as compared to those who have non-optimal plans. 
• Customer segmentation and strategy development based on customer lifetime value: A case study, Kim, S. Y., Jung, T. S., Suh, E. H., & Hwang, H. S. (2006). Expert systems with applications, 31(1), 101-107. An effective marketing strategy has a great impact on building a long term relationship with the customers. Then, the customers are segmented on the basis of their value. The authors present a structure to evaluate customer value and customer segmentation in the wireless telecom industry. Class imbalance is an important concept of data mining. 
• Handling class imbalance in customer churn prediction, Burez, J., & Van den Poel, D. (2009). Expert Systems with Applications, 36(3), 4626-4636. This paper investigates the class imbalance to calculate the churn prediction. The authors use proper metric, including Lift and AUC to boost the sampling performance. AUC is independent of a threshold and can give a good accuracy rate. Lift is also used for accuracy, but its primary purpose is marketing. WRF(Weighted Random Forests) is a cost related learner. Its comparison is made with the Logistic Regression. 
• Defection detection: Measuring and understanding the predictive accuracy of customer churn models, Neslin, S. A., Gupta, S., Kamakura, W., Lu, J., & Mason, C. H. (2006). Journal of marketing research, 43(2), 204-211. This paper analyses the factors affecting the accuracy of churn prediction. A tournament was held on which practitioners and students made downloads from a government website. They tried to predict on the basis of 2 databases of validation. There were many differences found with respect to performance. Their implications have been discussed thoroughly. 
• Customer churn analysis: Churn determinants and mediation effects of partial defection in the Korean mobile telecommunications service industry, Ahn, J. H., Han, S. P., & Lee, Y. S. (2006). Telecommunications policy, 30(10-11), 552-568. This study focuses on how to retain customers in the telecom industry. The factors of customer churn have been examined in the telecom sector of Korea. The findings are that the call quality affects the churn directly or indirectly. However, users who avail card plans also have high propensity to churn. The change in the status of a user can change the churn rate.
• Customer churn prediction in telecommunications, Huang, B., Kechadi, M. T., & Buckley, B. (2012). Expert Systems with Applications, 39(1), 1414-1425. This paper analyses the churn prediction for landline customers. The authors apply 7 prediction methods (Linear Classification, Decision Trees, Support Vector Machines, Naive Bayes, Evolutionary Data Mining Technique, Logistic Regression, and Multi-Layer Perceptron Neural Networks). This is more helpful in retaining existing clients.
• Applying Bayesian Belief Network approach to customer churn analysis: A case study on the telecom industry of Turkey, Kisioglu, P., & Topcu, Y. I. (2011). Expert Systems with Applications, 38(6), 7151-7157. Due to the competition in the Turkish telecom industry, user Churn has become a great problem. Applying the Bayesian Belief Network (BBN), the visitors propensity to churn can be estimated. The authors use a Chi-squared Automatic Interaction Detector (CHAID) to convert the discrete variables into continuous. The experts opinion, correlation analysis and multicollinearity tests help in identifying the variables affecting the churn rate.
• The effects of customer satisfaction and switching barrier on customer loyalty in Korean mobile telecommunication services, Kim, M. K., Park, M. C., & Jeong, D. H. (2004). Telecommunications policy, 28(2), 145-159. The mobile industry in Korea is focusing more on the strategies to retain the existing clients applying the promotional techniques of user loyalty than on making new clients. This paper highlights the ways to get the customers satisfaction and overcome the switching barrier.
• A customer churn prediction model in telecom industry using boosting, Lu, N., Lin, H., Lu, J., & Zhang, G. (2014). IEEE Transactions on Industrial Informatics, 10(2), 1659-1665. The CRM system (Customer Relationship Management) is an emerging trend in the telecom industry. The authors use a boosting algorithm to separate the visitors and make their 2 groups. There is a higher risk group. Using Logistic Regression (LR), the authors create a model of the Churn Prediction on both groups. With boosting, the Churn information is easy to separate to make a prediction analysis.