Comparative Performance of Supervised Machine Learning Models for Customer Churn Prediction Using E-Commerce Behavioral Data

Abstract

Customer churn remains a major challenge for e-commerce organizations because customer retention directly affects profitability and business sustainability. This study presents a comparative evaluation of five supervised machine learning algorithms for customer churn prediction using 3,000 customer records obtained from an e-commerce dataset on Kaggle, comprising two classes: churned and non-churned customers. The results showed that XGBoost achieved the best overall performance, with 95.67% accuracy, 98.59% precision, 97.88% recall, 97.73% F1-score, and 97.29% AUC. Gradient Boosting and SVM followed with F1-scores of 95.00% and 94.33%, respectively, while Random Forest achieved 94.17%. Decision Tree recorded the lowest AUC of 80.94%. The findings demonstrate that XGBoost was the most effective algorithm for customer churn prediction under the experimental conditions of the study. Keywords: Customer Churn, E-commerce, Machine Learning, XGBoost