A Cost-Sensitive Ensemble Model for e-Commerce Customer Behavior Prediction with Weighted SVM

Abstract

In today’s highly competitive environment, effective customer relationship management (CRM) is critical for every company, especially e-commerce businesses. Analyzing customer behavior is an initial step in marketing strategies and revenue generation, and then companies can predict purchasing behavior to enhance efficiency and boost profits. Many practitioners have noted that in customer behavior prediction, some special requirements, such as minimizing false predictions, need to be implemented to prevent the churn of potential customers. On the other hand, many studies on customer behavior prediction do not leverage unstructured data, such as conversation records, to improve a prediction model’s performance. We propose a prediction model based on weighted support vector machines (SVM) that can reduce the churn of customers while maintaining reasonable accuracy. We validate our approach using customer data from an e-commerce company and conduct multi-dimensional comparisons to demonstrate the practicality of our method.