Using Deep Learning to Predict Customer Churn in a Mobile Telecommunication Network

Customer churn is defined as the loss of customers because they move out to competitors. It is an expensive problem in many industries since acquiring new customers costs five to six times more than retaining existing ones [1-4]. In particular, in telecommunication companies, churn costs roughly $10 billion per year [5]. A wide range of supervised machine learning classifiers have been developed to predict customer churn [6-9]. In general, these models base their effectiveness in the feature engineering process which is usually time consuming and thus tailored to specific datasets. Since deep learning automatically comes up with good features and representation for the input data; we investigated the application of autoencoders, deep belief networks and multi-layer feedforward networks with different configurations. We report results for predicting customer churn using a four-layer feedforward architecture. To scale the model to full-sized high dimensional customer data, like the social graph of a customer, we introduced a data representation architecture that allows efficient learning across multiple layers of detailed user behavior representations. In this article, we use billions of call records from an enterprise business intelligence system and present our current work towards using deep learning for predicting churn in a prepaid mobile telecommunication network. To the best of our knowledge this is the first work reporting the use of deep learning for predicting customer churn. On average, our model achieves 77.9% AUC on validation data, significantly better than our prior best performance of 73.2% obtained with random forests and an extensive custom feature engineering applied to the same datasets.