Enhancing Telecom Customer Retention through Data Mining-Based Churn Prediction

Abstract

Customer retention remains a significant challenge for telecom service providers, with churn rates directly impacting revenue and customer acquisition costs. Traditional machine learning approaches have been widely used for churn prediction, but data mining techniques offer alternative insights by uncovering hidden patterns in customer behavior. This study explores clustering, association rule mining, and anomaly detection as key data mining approaches for churn prediction. Unlike previous studies that rely on supervised learning models, this research leverages unsupervised and rule-based techniques to provide interpretable insights. The dataset, sourced from a telecom service provider, includes customer demographics, usage behaviour, billing details, and complaint records. Experimental results demonstrate that clustering methods such as K-Means and DBSCAN effectively segment high-risk customers, association rule mining identifies key service combinations linked to churn, and anomaly detection methods highlight outliers with high churn probability. The findings suggest that integrating data mining approaches enhances customer retention strategies and provides telecom companies with proactive decision-making tools. Future research should focus on real-time churn detection and hybrid predictive models.