Churn factors identification from real-world data in the telecommunications industry: case study

Abstract

The goal of this study is to present high quality features usable for modeling the churn phenomenin in the telecommunications industry. This paper presents the process of identifying churn factors based on real-world data coming from one the biggest Polish telecommunications operator. The feature extraction and construction techniques are used to create preliminary sets of variables potentially describing the churn phenomenon. Variables describing subscriber demographics, contract conditions and contract usage details have been extracted for two consequent six-month periods. In the next steps, correlation and collinearity analysis were employed in order to select the best variables – candidates for churn prediction. As a final result only most promising variables were presented. Those results can be used to extend data gathered by telecommunications providers, and enhance the set of features applied in churn modeling.