A Machine Learning Approach: Predicting Customer Churn in Broadband Internet Services for a Small-Scale Telecommunications Company in Sri Lanka

Abstract

"Customer churn, also known as customer attrition is a major problem that telecommunication service providers in particular face in today’s rapidly growing and competitive telecom landscape in Sri Lanka. Techniques for predicting customer churn in the broadband service field in Sri Lanka have not been extensively studied. Currently, either very little churn prediction has been conducted for fixed-line networks offering broadband internet services, or the literature on churn prediction in telecommunications does not provide detailed methodologies for using information in churn prediction for a specific product. The problem statement in this research revolves around identifying the factors and patterns that contribute to customer churn in the context of a small-scale telecommunications company providing Broadband internet services.

The models that were picked to train the dataset are Logistic Regression, SVM, Naïve Bayes, KNN, Decision Tree, and Random Forest. All of these falls under the classification algorithm type since it is established that the research problem is a classifying problem. It was concluded that the SVM model has superior performance metrics due to it producing the highest Accuracy and Precision rates of 87.2% and 88% respectively and having the best predictability in comparison to the metrics across the board. Additionally, according to the key findings of the study, Sri Lankan Internet subscribers are particularly price sensitive, with the majority of users likely to switch from one network to another due to reduced Value-Added Service rates offered by another company for their Broadband services."