Abstract
Several machine learning models have been proposed to address customer churn problems. In this work, the authors used a novel method by applying deep convolutional neural networks on a labeled dataset of 18,000 prepaid subscribers to classify/identify customer churn. The learning technique was based on call detail records (CDR) describing customers activity during two-month traffic from a real telecommunication provider. The authors use this method to identify new business use case by considering each subscriber as a single input image describing the churning state. Different experiments were performed to evaluate the performance of the method. The authors found that deep convolutional neural networks (DCNN) outperformed other traditional machine learning algorithms (support vector machines, random forest, and gradient boosting classifier) with F1 score of 91%. Thus, the use of this approach can reduce the cost related to customer loss and fits better the churn prediction business use case.
Highlights
During the last decade competition became a real concern for telecommunication providers (Bin et al, 2007)
We have demonstrated how deep learning algorithms can approach the churn business problem by applying structural data from a real mobile telecom provider to different deep convolutional neural networks (DCNN) models
We have shown that our deep learning models achieve good results in the churn prediction business problem by surpassing the traditional machine learning algorithms in term of AUC, F1 Score, precision, and recall
Summary
During the last decade competition became a real concern for telecommunication providers (Bin et al, 2007). Previous studies have shown that different state of art predictive models were used to predict the churning problem by (training binary classifiers) using labeled churner/non churner dataset involving the Traditional hand-crafted features (Keramati et al, 2014; Vafeiadis et al, 2015; The Chartered Institute of Marketing, 2010) or social network analysis technique (Phadke et al, 2013; Richter et al, 2010).motivated by the recent advances of deep learning (LeCun et al, 2015) in different area of research, we propose a novel method using convolutional neural network for customer’s churn use case. We found that ConvNets provides more meaningful and useful representations (yielded to optimal results), outperforming other conventional machine learning algorithms such as Naïve Random Forest, Gradient Boosting Classifier, and Support Vector Machine This result indicates that our approach represents an important contribution to one open industrial question: how deep learning can be a useful method in addressing issues related to business telco data?.
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
