Enhanced Customer Churn Prediction in Telecommunications: A Multi-Dataset Comparative Study of Ensemble Machine Learning Models with Explainable AI

Kyaw, Hayman (2025) Enhanced Customer Churn Prediction in Telecommunications: A Multi-Dataset Comparative Study of Ensemble Machine Learning Models with Explainable AI. Masters thesis, Dublin, National College of Ireland.

Preview	PDF (Master of Science) Download (1MB) | Preview
Preview	PDF (Configuration Manual) Download (296kB) | Preview

Abstract

Customer churn prediction is a critical issue in telecommunications, as it is with five to ten times higher costs of acquiring new customers than retaining existing ones. This study addresses the gap between the prediction accuracy and business interpretability based on a comparative study of traditional machine learning algorithms (Logistic Regression, Random Forest) and advanced ensemble algorithms (XGBoost, LightGBM) on three varied telecommunications datasets consisting of 158,090 customer records.

In this study, explainable AI was utilized by combining model-specific interpretability methods with LIME (Local Interpretable Model-agnostic Explanations). This took complex predictions and turned them into clear, actionable insights for the business. It was shown that LightGBM produced the most stable performance in all datasets with an average AUC-ROC of 0.7578, and gradient boosting techniques significantly worked well compared to other traditional techniques. There was a significant difference between the performance of different datasets, with Dataset 2 showing the highest AUCROC of 0.9062 and Dataset 1 with the lowest possible performance of 0.6803, proving that the dataset characteristics are the key factors that determine achievable accuracy.

Interpretability analysis found out that equipment age and tenure are the key drivers of churn, with the customers whose equipment is above 515-530 days of age or have a tenure below 9 months of equipment having a very high risk. The patterns of contract type and usage became context-related forces. The consistency of rankings of feature importance among various types of models confirmed the usefulness of identified churn patterns revealing that advanced ensemble techniques with explainable AI can offer both high predictive power and transparency needed by business decision-making in telecommunication customer retention models.

Item Type:	Thesis (Masters)
Supervisors:	Name Email Simiscuka, Anderson UNSPECIFIED
Subjects:	Q Science > QH Natural history > QH301 Biology > Methods of research. Technique. Experimental biology > Data processing. Bioinformatics > Artificial intelligence Q Science > Q Science (General) > Self-organizing systems. Conscious automata > Artificial intelligence H Social Sciences > HF Commerce > Marketing > Consumer Behaviour Q Science > Q Science (General) > Self-organizing systems. Conscious automata > Machine learning H Social Sciences > HD Industries. Land use. Labor > Specific Industries > Telecommunications Industry
Divisions:	School of Computing > Master of Science in Artificial Intelligence for Business
Depositing User:	Ciara O'Brien
Date Deposited:	03 Sep 2026 08:25
Last Modified:	03 Sep 2026 08:25
URI:	https://norma.ncirl.ie/id/eprint/9779

Actions (login required)

View Item