Churn Prediction of Fiber to Home Users

Abstract

Due to the competitive strategies of service providers, Customers actively swap from one service provider to another to satisfies their needs which triggers them to ‘churn’. Hence, It’s time for the telecommunication industry to make necessary predictive decisions to increase the ‘happiness factor of customers’ and ‘Retaining the customers with stabilizing their market value.’This study discusses the strategies that can predict the Churn and the remedies of minimizing the churn factor of the fiber to home users. That could help relevant stakeholders to take a suitable decision at the right time and mitigate customers from churning, thereby reducing the churn rate. 14,461 instances related to a telecom sector are used in this study with fourteen network-related feature parameters for developing a churn prediction model for fiber to the home users, in which some of the features are numerical and some of the features are categorical. Among these features, eight significant features are selected using the Recursive Feature Elimination Technique. These variables are used as input variables for Logistic Regression classifier to build a churn prediction model. Model performance is measured using Recall, Precision, F1-score, Support, confusion Matrix, and ROC curve. It was able to achieve significant accuracy from the model with 0.86 area under the curve. Market research survey was carried out to observe why people are unhappy with fiber to home services. These results are used to develop business and technical strategies to retain the fiber to the home customers.