Customer Churn Prediction in the Telecommunications Industry Using Computational Methods

Introduction Customer churn remains a major challenge for telecommunications companies as competition intensifies and customer expectations increase. Developing accurate and interpretable churn prediction models is essential for identifying at-risk customers and designing effective retention strategies. This study investigates churn prediction using two machine learning approaches, K-Nearest Neighbors (KNN) and Decision Trees, focusing on both model performance and interpretability. Methods The analysis was conducted using the Customer Churn Prediction ۲۰۲۰ dataset, containing ۴,۲۵۰ customer records and ۲۰ features. Two main algorithms were implemented. KNN performance was evaluated using three validation strategies: Train-Test Split, Leave-One-Out Cross-Validation (LOOCV), and ۱۰-Fold Cross-Validation. Decision Tree models were built using the CART algorithm with entropy-based splitting. Both R and Python were used to generate decision rules, assess feature importance, and compare model behavior. Results KNN achieved relatively stable accuracy across all validation methods (۰.۸۶۰–۰.۸۶۸) but demonstrated low sensitivity in detecting churn cases. LOOCV had the lowest sensitivity (۰.۰۵) but the highest specificity (۰.۹۹۸), while Train-Test Split and Batch CV showed moderately higher sensitivity (~۰.۱۳). Decision Trees outperformed KNN in interpretability and provided actionable insights. Thirteen key rules were extracted, highlighting total daily call duration and the number of customer service calls as the strongest churn predictors. Discussion & Conclusion The findings show that although KNN performs well in overall accuracy, it struggles with identifying churned customers. Decision Trees offer clearer interpretability and better support practical retention decisions. Their extracted rules help telecom operators understand customer behavior patterns and design targeted interventions. Overall, interpretable models such as Decision Trees are more suitable for effective churn management and strategic decision-making in the telecommunications sector

کلیدواژه ها:

Machine Learning ، Customer Churn ، Telecommunications Industry ، K-Nearest Neighbors (KNN) ، Decision Tree ، Feature Importance ، Cross-Validation ، Model Interpretability.