Effectual predicting telecom customer churn using deep neural network

Abstract

Telecom industry has seen a phenomenal growth throughout the world in recent times. Today companies in this sector are putting their best efforts to retain their churning customers by satisfying them with offers and discounts. It is due to the fact that, acquiring a new customer is far more expensive than retaining an existing one. Deep neural network learns on its own in a supervised manner and thus can be used in this regard efficiently. In this paper we have used the H2o package of deep learning to predict telecom customer churn. H2o package stem from a multi-layer artificial neural network. Number of hidden layers, epoch, number of neurons, hidden dropout ratio, input dropout ratio and activation function have been varied to achieve high sensitivity value. Sensitivity is the percentage of churners who are correctly predicted as churning customers. Our model has achieved sensitivity of 85% and thus the results are satisfactory.