Bayesian network approach to predict mobile churn motivations: Emphasis on general Bayesian network, Markov Blanket, and what-if simulation

Abstract

As mobile telecommunication service becomes indispensable to our daily life, predicting the reasons of mobile churn seems essential from the perspective of mobile service providers. Previous studies have been focused on mobile churn prediction itself, not churn motivations which can play as a good indicator to forecasting real churn. Therefore, main focus of this study is placed on predicting mobile churn motivations, instead of mobile churn prediction. We propose BN approach to predict mobile churn motivation, adopting three types of BN models such as Naïve BN (NBN), Tree Augmented NBN (TAN), and General BN (GBN). To prove its validity in predicting mobile churn motivations, benchmarking classifiers were adopted and their performance was compared with BN classifiers. Through analyzing the empirical results, we found three advantages of GBN-(1) GBN performance is competitive compared with other benchmarking classifiers, (2) Markov Blanket (MB) variables are considerably small in number and make it handy for decision makers, and (3) what-if simulation is possible, which is not possible in other benchmarking classifiers. Practical implications of empirical results were addressed. © 2010 Springer-Verlag Berlin Heidelberg.