HPMANN: Hyper Parameterized Modified Artificial Neural Networks for Predicting Churn Customers in Telecom Industry

Abstract

Customer churn is a prominent factor in the economical marketplace and it is exceedingly studied in bank customer data, insurance data, and the telecom sector. It is those particular clients who have chosen to leave the organization and moving to the next rival in the marketplace. Companies are seeking to develop means to predict potential customers to churn to improve their business. Existing techniques do not show that much performance. So, require a model to improve churn prediction performance. In this work, Hyper Parameterized Modified Artificial Neural Networks (HPMANN) algorithm is introduced to increase the churn prediction performance considerably. This research contains three modules. The first module is the pre-processing module, traditional mechanisms are applied to complete the data cleaning and data transformation process. In the second module, Feature selection is performed using a modified BAT optimization algorithm. In the third module, classification is performed by using HPMANN, all the previous studies considered the parameters of neural networks on a random basis or by using brute force techniques. To accomplish a standard mechanism, this research designed a Cross-Validation Optimization (CVOPT) algorithm to find the best values for performing parameters turning. These best values are passed as the hyper parameter values of our customized neural network. Hence the evaluation parameters show that the proposed HPMANN approach gives better churn prediction performance than existing techniques