A Customer Churn Prediction Using CSL-Based Analysis for ML Algorithms: The Case of Telecom Sector

Abstract

The loss of customers is a serious issue that needs to be addressed by all major businesses. Companies, especially in the telecommunications industry, are trying to find ways to predict customer churn because of the direct impact on revenue. Therefore, it is important to identify the causes of customer churn to take measures to decrease it. Customer churn occurs when a company loses customers because of factors such as the introduction of new offerings by rivals or disruptions in service. Under these circumstances, customers often decide to end their subscription. Predicting the likelihood of a customer defecting by analyzing their past actions, current circumstances, and demographic data is the focus of customer churn predictive modeling. Predicting customer churn is a well-studied problem in the fields of data mining and machine learning. A common method for dealing with this issue is to employ classification algorithms to study the behaviors of both churners and non-churners. However, the current state-of-the-art classification algorithms are not well aligned with commercial goals because the training and evaluation phases of the models do not account for the actual financial costs and benefits. Different types of misclassification errors have different costs, so cost-sensitive learning (CSL) methods for learning on data have been proposed over the years. In this work, we present the CSL version of various machine learning methods for Telecom Customer Churn Predictive Model. Furthermore, also adopted feature selection strategies along with CSL in real-time telecom dataset from the UCI repository. The proposed combination of CSL with ML, the results outperforms the state-of-the-art machine learning techniques in terms of prediction accuracy, precision, sensitivity, area under the ROC curve, and F1-score.