Modeling Customer Churn Behavior in E-commerce Using Bayesian Networks

Abstract

Recent studies have reported that companies’ efforts to keep their current customers are considerably less costly than acquiring new ones. Therefore, it is critical for companies to analyze and understand the loss of customers and avoid the mistakes made throughout the customers’ journey. Both in theory and practice, churn analysis is widely used in industries such as telecommunications, insurance, and financial services since customers might switch to competitors easily. Limited research has been conducted on churn analysis and prediction in e-commerce. In this study, the factors, directly and indirectly, affecting the loss of customers in e-commerce are discussed, and an accurate and effective churn prediction model is suggested. The proposed probabilistic graphical model based on Bayesian networks allows (i) representing the relationships among factors in a convenient graphical representation (ii) both deductive and abductive reasoning under uncertainty, (iii) examining several different what-if scenarios, (iv) predicting consequences of possible interventions, and (v) using expert views and if available historical data. The applicability and effectiveness of the model are presented through an application. Analysis with actual data demonstrated that the proposed model achieves high prediction accuracy.