Predicting Customer Churn in Telecom Sector based on Penalization Techniques and Ensemble Machine Learning

  • Abstract
  • Keywords
  • References
  • PDF
  • Abstract

    Customer Churn Prediction model (CCP)aims to detect customers with a high propensity to leave. The target of this research is to handle a large scale Telecommunication Company and identify potential churn. In the proposed research, Predictive Mean Matching (PMM) algorithm used to handle missing values, instead of removing features or observations with high missing data.

    First Ensemble Machine learning classifieris offered to investigate and compare the combining of an Ensemble learner based on Generalized Linear Model (GLM) and the prediction values based on tree model using a Random Forest classifier. The suggested CCP model employed the Weighted Accuracy and Diversity (WAD) as an algorithm to find the optimal weights for the proposed Ensemble classifier.

    The second Ensemble learner based on the generalized linear model is incorporated of penalized methods (Ridge, Lasso,and ElasticNet) with a Logistic Regression method on the binomial family. Randomly generate values between [0, 1] became the weights for this classifier. The Weights are selected according to the principle that weights of highervalue are assigned for great performance classifier to ensure the highestaccuracy of Churn Prediction model. 10-fold, based on five times repeated Cross-Validation (CV) performance technique used to enable efficient and automatic search for the optimal value of lambda λ parameter for penalization methods.

    The two Ensemble classifiers incorporated within a customer churn prediction model to handle a large scale dataset, time-dependent features label, and an imbalance data distribution in the Telecommunication industry.

    Experimental results show an increasein predictive performance. In addition, the results depicted that using of ensemble learning has brought a significant improvement for individual base learners in terms of performance indicators such as Area under Curve (AUC), sensitivity, specificity, Accuracy, and Mean Square Error(MSE), Accuracy is the best candidates for churn prediction tasks.



  • Keywords

    Customer Churn Prediction, Random Forests, Ensemble Machine Learning, Weighted accuracy and diversity, Telecommunication Industry, Boosting, Penalization Method, Regularization Techniques.

  • References

      [1] J. Donald, “Predicting Attrition in Financial Data with Machine Learning Algorithms,” 2018.

      [2] M. K. Sahu, R. Pandey, and S. Silakari, “ISSN NO : 0076-5131 Analysis of Customer Churn Prediction in Telecom Sector Using Logistic Regression and Decision Tree Keywords :” J. Appl. Sci. Comput., vol. 5, no. 6, pp. 62–67, 2018.


      [4] G. C. Esteves, “Churn Prediction in the Telecom Business,” p. 96, 2016.

      [5] W. Verbeke, “Profit-driven data mining in massive customer networks: new insights and algorithms,” no. 379, 2012.

      [6] G. Vink, L. E. Frank, J. Pannekoek, and S. van Buuren, “Predictive mean matching imputation of semicontinuous variables,” Stat. Neerl., vol. 68, no. 1, pp. 61–90, 2014.

      [7] H. Abbasimehr, M. Sestak, and M. J. Tarokh, “A comparative assessment of the performance of ensemble learning in customer churn prediction.,” Int. Arab J. Inf. Technol., vol. 11, no. 6, pp. 599–606, 2014.

      [8] G. Louppe, “Understanding Random Forests: From Theory to Practice,” 2014.

      [9] K. Bailey, J. Miller, and Valerie Santiago-Gonzalez, “predicting diabetes diagnosis in African Americans using Ensemble machine learning.”

      [10] I. Stephen Nabareseh, “Predictive analytics: a data mining technique in customer churn management for decision making Prediktivní analytika: technika data miningu pro rozhodování s využitím v řízení odchodu zákazníků,” no. February, 2017.

      [11] F. Andreis, “Shrinkage methods (ridge, lasso, elastic nets),” no. November 2017.

      [12] J. Vorlíčková, “Least Absolute Shrinkage and Selection Operator Method,” 2017.

      [13] A. Agarwal, G. Verma, H. B. Sri, K. Mannem, and F. Hamid, “Indian Institute of Technology, Kanpur Department of Industrial and Management Engineering IME672A Data Mining and Knowledge Discovery Course Project Report,” 2016.

      [14] G. Vink, G. Laserdisc, and S. Van Buuren, “Partitioned predictive mean matching as a multilevel imputation technique,” Psychol. Test Assess. Model. vol. 5, no. 4, pp. 1–16, 2015.

      [15] P. Allison, “Imputation by Predictive Mean Matching: Promise & Peril,”, 2015. [Online]. Available:

      [16] A. J. van der Koij, “Regularization with ridge penalties, the lasso, and the elastic net for regression with optimal scaling transformations,” Predict. Accuracy Stab. Regrets. With Optim. Scaling Transform. no. 2006, pp. 65–90, 2007.

      [17] J. Lanford, T. Nykodym, A. Rao, and A. Wang, Generalized Linear Modeling with H2O’s R Package. 2015.

      [18] S. Dardouri and R. Bouallegue, “Performance Analysis of Regularized Linear Regression Models For Oxazolines and Oxazoles Derivatives Descriptor Dataset,” vol. 1, no. 4, pp. 111–123, 2013.

      [19] D. Dalpiaz, R for Statistical Learning. 2017.

      [20] Art Owen, “Regularization : Ridge Regression and the LASSO the Bias-Variance Tradeoff,” 2007.

      [21] E. Krona, “A simulation study of model fitting to high dimensional data using penalized logistic regression Mathematica institution,” Stockholm University.

      [22] D. S. De Groot, “Churn prediction in telecommunication Classification problem,” 2017.

      [23] A. Lemmens and C. Croux, “Bagging and Boosting Classification Trees to Predict Churn,” J. Mark. Res., vol. 43, no. 2, pp. 276–286, 2006.

      [24] J. Van Haver, “Benchmarking analytical techniques for churn modeling in a B2B context,” 2016.

      [25] C. Zhang and Yunqian Ma Editors, Ensemble Machine Learning. 2012.

      [26] J. Vijaya and E. Sivasankar, “Computing efficient features using rough set theory combined with ensemble classification techniques to improve the customer churn prediction in the telecommunication sector,” Computing, vol. 100, no. 8, pp. 839–860, 2018.

      [27] X. Zeng, D. F. Wong, and L. S. Chao, “Constructing better classifier ensemble based on weighted accuracy and diversity measure,” Sci. World J., vol. 2014, 2014.

      [28] M. Ewing, “Teknisk-naturvetenskaplig fakultet UTH-enheten”, 2012.




Article ID: 27977
DOI: 10.14419/ijet.v7i4.19.27977

Copyright © 2012-2015 Science Publishing Corporation Inc. All rights reserved.