<![CDATA[As the number of e-commerce users continues to expand, analyzing user activity has become essential for improving service quality, since customers are a key factor in strengthening competitiveness and business performance in this industry. Churn, in which customers discontinue service due to dissatisfaction, requires accurate detection to support effective retention strategies. To improve churn prediction accuracy, this study explores the optimal combination of feature selection methods and machine learning classifiers. ANOVA F-Test, Information Gain, Chi-Square, Recursive Feature Elimination (RFE), Variance Threshold, and Least Absolute Shrinkage and Selection Operator (LASSO) are the six feature selection techniques that are assessed in this study. These are combined with three classification algorithms: Gradient Boosting (GB), Random Forest (RF), and Logistic Regression (LR). Gradient Boosting with either Information Gain or LASSO yields the best results, achieving approximately 98% accuracy, 97% precision, 92% recall, 96% F1-score, and 99% ROC-AUC, based on testing on a publicly available Kaggle dataset. This approach outperforms prior studies, providing a robust framework for e-commerce developers to anticipate customer loss. By identifying key predictors of churn, the proposed model offers actionable insights to enhance service and enable proactive customer management.]]>
Copyrights © 2026
