A systemic method of nesting multiple classifiers using ensemble techniques for telecom churn prediction

Abstract

In this contemporary world, almost every business and companies deploy machine learning methods for taking exemplary decisions. Predictive Customer analytics supports to accomplish of momentous insights from customer data. This trend is more distinct in the telecommunication industry. The most challenging issue for telecom service providers is the increased churn rate of customers. In Existing works, combining various classification algorithms to design hybrid algorithms as well as ensembles have reported best results compared to single classifiers. But, selecting classifiers for creating an effective ensemble combination is challenging and are still in investigation. This work presents a various type of ensembles such as Bagging, Boosting, Stacking and Voting to combine with different Base classifiers in a systemic way. Experiments were conducted with benchmark Telecom customer churn UCI dataset. It is inferred that; Ensemble learners outperform single classifiers due to its strong classifying ability. The models designed were affirmed using standard measures such as AUC, Recall, F-Score, TP-rate, Precision, FP-Rate and Overall Accuracy. This way of combining multiple classifiers as an ensemble achieves the highest accuracy of 97.2% from bagged and boosted-ANN.