Scholarlink

<p>In the competitive e-commerce landscape of today, understanding and predicting customer churn is critical for optimizing business strategies. This study proposes a machine learning-based framework to forecast customer churn and support targeted retention efforts. Using a real-world dataset containing demographic, transactional, and behavioral data, key churn indicators were identified. Customers were segmented using the K-Prototypes clustering algorithm, enabling the classification of groups such as high-value loyal customers and at-risk buyers. RFM, which stands for Recency, Frequency, and Monetary analysis, was also incorporated to assess customer lifetime value. Three classification models—Logistic Regression, Decision Tree, and Random Forest—were developed and compared. Logistic Regression outperformed the others, achieving a Recall of 91.51% and an F1 Score of 67.60%, with a predicted churn rate of 74.5%, which closely matches the actual churn rate of 63%. Additionally, two new models, Naive Bayes and SVM, were tested with k-fold cross-validation, with Naive Bayes achieving the highest Recall (94.49%) and F1-Score (67.79%) among all models tested. The proposed hybrid framework integrates behavioral clustering and predictive modeling, offering actionable insights for churn mitigation. Future work may enhance this approach by incorporating deep learning techniques and external contextual variables such as economic indicators or seasonal effects.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Machine learning-based prediction and optimization of customer churn in e-commerce

  • 期刊论文

  • 发表日期：

    2026年5月18日

查看全文
  • Bao Tran Giang,
  • Ngoc Diem Thuy Nguyen,
  • Thi Ngoc Chau Bui,
  • Thuy Chi Nguyen,
  • Duy Thanh Tran

摘要

In the competitive e-commerce landscape of today, understanding and predicting customer churn is critical for optimizing business strategies. This study proposes a machine learning-based framework to forecast customer churn and support targeted retention efforts. Using a real-world dataset containing demographic, transactional, and behavioral data, key churn indicators were identified. Customers were segmented using the K-Prototypes clustering algorithm, enabling the classification of groups such as high-value loyal customers and at-risk buyers. RFM, which stands for Recency, Frequency, and Monetary analysis, was also incorporated to assess customer lifetime value. Three classification models—Logistic Regression, Decision Tree, and Random Forest—were developed and compared. Logistic Regression outperformed the others, achieving a Recall of 91.51% and an F1 Score of 67.60%, with a predicted churn rate of 74.5%, which closely matches the actual churn rate of 63%. Additionally, two new models, Naive Bayes and SVM, were tested with k-fold cross-validation, with Naive Bayes achieving the highest Recall (94.49%) and F1-Score (67.79%) among all models tested. The proposed hybrid framework integrates behavioral clustering and predictive modeling, offering actionable insights for churn mitigation. Future work may enhance this approach by incorporating deep learning techniques and external contextual variables such as economic indicators or seasonal effects.