CRBT customer churn prediction: can data mining techniques work?

Abstract

Coloring Ring Back Tone (CRBT) is one of the most successful Value-added (VAD) services in China telecommunication operators. Under fierce competition conditions, CRBT customer churn has significantly decreased the profits of operators. Thus churn management has become a major focus to retain subscribers via satisfying their needs under resource constraints. One of the challenges is that churn prediction specific to this business is not available in existing literature. Through empirical evaluation, this study analyse the features of CRBT, compare various data mining techniques that can assign a 'propensity to churn' to each CRBT subscriber. The results indicate that our models can achieve satisfactory prediction effectiveness by using customer demographics, billing and service usage information. At the same time, we find some new symptoms different from existing telecom churn literature, and try to explain them, and point out which predictors are needed to intensively monitor by telecom operators.