Saved in:
Bibliografiske detaljer
Hovedforfatter: Dadakidis, Giorgos
Format: Recurso digital
Sprog:engelsk
Udgivet: Zenodo 2025
Fag:
Online adgang:https://doi.org/10.5281/zenodo.15978511
Tags: Tilføj Tag
Ingen Tags, Vær først til at tagge denne postø!
Indholdsfortegnelse:
  • <p>Customer churn is a persistent and costly challenge for companies in the <br>telecommunications sector, where maintaining existing subscribers is often more <br>profitable than acquiring new ones. Accurately identifying customers who are likely to <br>leave is critical for enabling targeted retention strategies. However, churn prediction is <br>complicated by significant class imbalance, as the number of churners typically <br>represents a small fraction of the overall customer base.</p> <p>This thesis explores the application of machine learning techniques to the churn <br>prediction problem using a structured experimental approach. Five experimental settings <br>were designed to evaluate and improve model performance under imbalanced data <br>conditions: a baseline scenario using the original dataset, a cost-sensitive learning setup <br>with class weighting, a recall-optimized configuration through hyperparameter tuning, an <br>experiment incorporating synthetic oversampling (SMOTE) and a final experiment using <br>the top 20 important features . A variety of classification models were assessed, including <br>both traditional machine learning algorithms and neural networks. </p> <p><br>The study aims to investigate how different learning strategies and evaluation criteria <br>affect model behavior and performance in the context of churn prediction. Emphasis is <br>placed on addressing the imbalance issue, optimizing recall of the minority class, and <br>comparing the effectiveness of algorithmic and data-driven solutions. The findings provide <br>insights into the trade-offs and considerations involved in developing fair and practical <br>predictive models for real-world customer churn scenarios. </p>