•  
  •  
 

Corresponding Author ORCID Identifer

Lew Sook Ling https://orcid.org/0000-0003-4545-1163
Ram Avtar https://orcid.org/0000-0003-3653-5771

Section

Regular Issue

Abstract

Customer churn rate is one of the main challenges that e-Commerce businesses must address, as it may affect revenue and long-term growth. In this paper, predictive analysis was used to identify potential customers who would be lost. It uses two different types of data to evaluate and compare the performance of five Machine Learning (ML) models, namely logistic regression, decision tree, support vector machine, gradient boost and random forest. Accuracy, precision, recall rate and F1 score are used as indicators of model performance. Among them, random forest achieved the highest performance across both datasets, demonstrating superior predictive power and consistency. By validating results on multiple datasets, this paper not only ensures reliability but also highlights the stability of churn prediction models across varying data conditions. The findings indicate that effective churn prediction can support proactive customer retention strategies, enabling e-Commerce businesses to minimize losses and promote sustainable growth. The use of two datasets represents a notable contribution, as it provides a stronger basis for assessing model generalizability and avoiding bias toward a single data source. The findings indicate that accurate churn prediction can support proactive retention strategies, enabling e-Commerce businesses to reduce churned customers, enhance loyalty, and maintain a competitive advantage. Overall, this paper underscores the value of data-driven approaches, demonstrates the benefits of employing multiple datasets for model validation, and provides actionable insights for businesses aiming to implement effective and sustainable customer retention strategies.

Download

Share

COinS