Machine learning analysis of telecom customer churn in California

Abstract

Customer churn is a significant concern in the telecommunications industry, particularly in highly competitive and price-sensitive markets such as California. Although machine learning methods are widely applied for churn prediction, their practical adoption is often limited by insufficient interpretability. This study develops an interpretable and business-oriented churn analysis framework by comparing multiple machine learning models, including logistic regression (LR), random forest (RF), support vector machine (SVM), and extreme gradient boosting (XGBoost), combined with shapley additive explanations (SHAP)-based explanation techniques. Using a real-world California telecom customer dataset, the framework identifies key churn determinants related to contract characteristics, pricing sensitivity, and customer service behaviors. Experimental results demonstrate that XGBoost achieves the best predictive performance among the evaluated models, while SHAP provides transparent explanations of feature contributions, highlighting the importance of contractual and pricing factors for customer retention strategies.