Predicting Customer Churn Using Artificial Neural Networks: A Data-Driven Approach to Enhance Customer Retention in E-Commerce

Abstract

Customer churn poses a major challenge in e-commerce, making accurate prediction crucial for retention strategies. This study applies Artificial Neural Networks (ANNs) to predict churn in e-commerce and subscription-based businesses using a dataset of 1,644 customer records. Key independent variables include delayed deliveries, frequent product returns, high cart abandonment, poor customer service, high prices, better deals from competitors, complicated return processes, hidden fees, negative reviews, and low engagement, while the dependent variable is customer churn (churned or retained). A Multilayer Perceptron (MLP) ANN model with hyperbolic tangent and Softmax activation functions was employed, achieving 75.7% accuracy in churn prediction. Findings reveal that hidden fees, complicated return processes, and low engagement are the strongest churn predictors. These insights enable businesses to implement personalized marketing, improved service policies, and proactive support strategies. Despite its effectiveness, the study acknowledges model interpretability issues and data imbalance challenges. Future research can explore explainable AI techniques, sentiment analysis, and enhanced data preprocessing to improve churn prediction accuracy.