Developing an ensemble approach for predicting customer churn in telecommunication industry

Abstract

Customer satisfaction and retention are key goals and, at the same time challenges, for most of the modern companies which try to keep up with the times. To identify and retain the customers who are most likely to ‘break ties’ with the company, the latter spend much financial and technological resources. Those include advanced machine learning algorithms for customer churn prediction. This thesis explores a number of different common ML algorithms, including logistic regression, support vector machines, decision tree, random forest and XGBoost, which predict customer churn in wireless telecommunication industry. To mitigate the risks of non-accurate predictions, an ensemble algorithm is developed based on the weighted voting approach. In this thesis the performance of ensemble algorithm will be compared to those of all above mentioned to rank them by prediction accuracy and choose the best-performing one.