Predicting Customer Churn in E-Commerce Using Machine Learning: A Comprehensive Approach

Abstract

Customer churn prediction is a critical challenge in e-commerce, as retaining existing customers is often more cost-effective than acquiring new ones. This study evaluates the effectiveness of three machine learning algorithms Logistic Regression (LR), Random Forest (RF), and Extreme Gradient Boosting (XGBoost) for predicting customer churn using a structured e-commerce dataset that integrates demographic, behavioral, engagement, and transactional features. A systematic preprocessing pipeline was implemented, including missing value imputation, categorical encoding, and feature scaling where appropriate. Model performance was assessed using Accuracy, Precision, Recall, F1-Score, and Confusion Matrix analysis. The results reveal a clear performance gap between linear and ensemble-based approaches. Logistic Regression achieved 77% accuracy but demonstrated limited recall for churned customers, indicating reduced sensitivity in identifying at-risk users. In contrast, Random Forest and XGBoost both achieved 92% accuracy and substantially improved recall and F1-Score for the churn class. Among the evaluated models, XGBoost showed the strongest overall performance, achieving the highest recall and lowest false negatives, making it particularly suitable for retention-focused applications. The findings confirm that ensemble and boosting-based models are more effective in capturing nonlinear interactions among multidimensional churn determinants in e-commerce environments. This study contributes empirical evidence supporting the adoption of advanced machine learning approaches for customer retention optimization and provides practical guidance for integrating churn prediction models into e-commerce platforms.