Applied AI for Telecom Revenue: Examining eXplainable AI in Customer Churn Prevention

Abstract

This master’s thesis presents a comprehensive study on applying artificial intelligence, particularly eXplainable AI (XAI), to predict customer churn in Telenor’s businessto- business (B2B) sector. The primary objective of this research is to develop an interpretable machine-learning model that accurately predicts customer churn and provides actionable insights for retention strategies. Given the challenge of imbalanced datasets, the study employs XGBoost, a robust machine learning algorithm, and compares its performance against logistic regression. Both models are evaluated on various metrics, focusing on precision and recall, given the minority class’s importance in churn prediction. To enhance interpretability, SHapley Additive exPlanations (SHAP) are integrated with the XGBoost model, enabling stakeholders to understand the decision-making process. The thesis details the data collection and preprocessing stages, highlighting issues such as insufficient historical data and missing values. Advanced feature engineering techniques, including one-hot encoding and transformations, are employed to refine the dataset. The research also addresses computational challenges, leveraging Telenor’s migration to cloud-based solutions to improve efficiency. The study identifies the best-performing models through rigorous testing and validation and demonstrates that the XGBoost model, combined with SHAP values, provides a balance of accuracy and interpretability. The findings suggest that this approach can significantly enhance Telenor’s ability to retain customers by offering clear, data-driven insights into churn dynamics. This work contributes to applied AI in telecom by providing a practical, interpretable solution for customer retention. It demonstrates the value of integrating advanced machine learning techniques with explainable AI methods. Future work is anticipated to expand on daily churn predictions and incorporate more comprehensive datasets to refine the model’s accuracy and utility further.