Churn Analysis On Imbalanced Dataset In A Telecom Sector

Authors

  • Umair Mansoor, V. Sivakumar , Vinesh Thiruchelvam

Abstract

There is a sense of high competitiveness in the telecom sector not only based on their provided
services but also on their customer base. In such environment every customer has a very strong significance
to these telecom service providers and churning of any single of them is very costly to these companies. In
this a predictive model has been developed which can provide timely information for such potential
churning customers and for that purpose three different data mining classifiers have been employed in this
study which are Decision Tree (DT), Naïve Bayes (NB) and Neural Network (NN). Along with that
objective, the system also considers the highly useful factors which are most crucial and common to these
churners and such features were selected by using highly sophisticated Boruta technique. The predictive
system is designed on a real-world dataset which is inclined towards the class imbalance. Hence, this issue
of class imbalance has also been catered in this study by incorporating Adaptive Synthetic Sampling
(ADASYN) and Synthetic Minority over Sampling Technique (SMOTE) because of such class imbalance
problem the prediction of the system can be significantly biased to the majority group. The result of the
study showed that the performance of the model was significantly improved by employing these abovementioned class balancing and feature selection techniques. The highest predictive performance was
achieved by the DT model based on the balanced dataset by ADASYN and with the selected features by
Boruta technique. As the accuracy and kappa value of the model was increased from 74.83% to 85.42% and
45.35% to 70.84% respectively. These results were extremely satisfactory in terms of the proposed
objectives and research scope. For the future study and much more improved predictive modelling, a large
dataset with more sophisticated features related to the customers demographics and socioeconomics will be
highly beneficial and such systems can be incorporated in various telecom organizations to control the
churning behaviour of their customers.

Published

2020-01-31

Issue

Section

Articles