ChurnNet: Deep Learning Enhanced Customer Churn Prediction in Telecommunication Industry

Customer turnover in the Telecommunication Industry (TCI) is a major problem since the income of the service provider is mostly dependent on the retention of current consumers. It is imperative for the service providers in this competitive market to address the issues of their current clients about their services since the cancellation of the services by the clients and moving to new service providers will not benefit the service provider. Many studies have been conducted in the framework of TCI to forecast customer churn; yet, following the performance assessment of these studies reveals sufficient space for improvement. Thus, in this work, we proposed a new customer churn prediction architecture called ChurnNet to forecast TCI customer attrition. To enhance the performance of our proposed ChurnNet, residual block, squeeze and excitation block, and spatial attention module is combined with 1D convolution layer. Residual block facilitates the solution of the vanishing gradient issue. The ChurnNet can grasp the interdependency between and inside the channels respectively by means of squeeze and excitation block and spatial attention module. The experiment runs on three publicly available datasets in order to assess the performance. Three data balancing methods—SMote, SMoteEN, and SMOTETomek—are used as the datasets feature notable class imbalance problems. ChurnNet outperformed the state-of- the-art and obtained 95.59%, 96.94%, and 97.52% accuracy on 3 benchmark datasets respectively along with 10-fold cross-valuation and following the demanding experiment.