A graph-based churn prediction model for mobile telecom networks

Abstract

With the ever-increasing demand to retain the existing customers with the service provider and to meet up the competition between various telecom operators, it is imperative to identify the number of visible churners in advance, arbitrarily in telecom networks. In this paper, we consider this issue as a social phenomenon introduced to mathematical solution rather than a simple mathematical process. So, we explore the application of graph parameter analysis to the churner behavior. Initially, we try to analyze the graph parameters on a network that is best suited for node level analysis. Machine learning and Statistical techniques are run on the obtained graph parameters from the graph DB to select the most significant parameters towards the churner prediction. The proposed novel churn prediction methodology is finally perceived by constructing a linear model with the relevant list of graph parameters that works in a dynamic and a scalable environment. We have measured the performance of the proposed model on different datasets related to the telecom domain and also compared with our earlier successful models. © Springer-Verlag 2012.