Automated Feature Selection and Churn Prediction using Deep Learning Models

---------------------------------------------------------------------***--------------------------------------------------------------------Abstract – In this competitive world, mobile telecommunications market tends to reach a saturation state and faces a fierce competition. This situation forces the telecom companies to focus their attention on keeping the customers intact instead of building a large customer base. According to telecom market, the process of subscribers (either prepaid or postpaid) switching from a service provider is called customer churn. Several predictive models have been proposed in the literature for churn prediction. The efficiency of any churn prediction model depends highly on the selection of customer attributes (feature selection) from the dataset for its model construction. These traditional methods have two major problems: 1) With hundreds of customer attributes, existing manual feature engineering process is very tedious and time consuming and often performed by a domain expert: 2) Often it is tailored to specific dataset, hence we need to repeat the feature engineering process for different datasets. Since deep learning algorithms automatically comes up with good features and representation for the input data, we investigated their applications for customer churn prediction problem. We developed three deep neural network architectures and built the corresponding churn prediction model using two telecom dataset. Our experimental results show that deep-learning based models are performing as good as traditional classification models, without even using the hand-picked features.