Customer Churn Prediction using Neural Networks and SMOTE-ENN for Data Sampling

Abstract

To survive in the telecommunications industry's severe competition and to keep existing loyal customers, predicting prospective churn consumers has become a critical task that may be accomplished with efficient predictive models. For many years, churn studies have been utilized to boost profitability and make customer-company relationships more sustainable. Customer churn is estimated using a multi-layer perceptron prediction model based on ANN. Furthermore, the suggested model manages the data's uneven class distribution using an advanced oversampling strategy called SMOTE-ENN based on K-Nearest Neighbors. The models accuracy was compared with and without SMOTE-ENN and the model using SMOTE-ENN showed better results.