Optimising Customer Churn Prediction in Telecommunications: A Comparative Analysis of Machine Learning Models Using Hierarchical Time Series Forecasting

Abstract

This thesis investigates the application of hierarchical time series forecasting to predict customer churn in the telecommunications industry. By employing advanced machine learning models such as SARIMA, Prophet, ETS, LSTM, and XGBoost, the study aims to enhance the accuracy and relevance of churn predictions. The research compares the effectiveness of these models, focusing on their ability to forecast churn counts rather than individual churn events. The findings indicate that Prophet and XGBoost offer superior performance across different levels of data aggregation. This study contributes to the field by demonstrating the benefits of hierarchical forecasting for strategic planning and resource allocation, providing valuable insights into customer retention strategies.