A Novel Model for Partial and Total Churn Prediction in E-Commerce

Abstract

Abstract The e-commerce market is a rapidly growing industry, with many companies entering the market to provide customers with easy access to a variety of products and services. However, with the increasing number of e-commerce sites, customers are now able to move their purchases from one site to another or split their purchases among multiple sites. This trend creates a challenge for companies, as acquiring new customers is more costly than retaining existing ones. The proposed model is used to predict customer churn in the e-commerce market. Customer churn refers to customers who stop using a particular product or service. The model uses a dataset from a B2C multi-category e-commerce application that describes customer behavior and interactions. The model defines and predicts the types of customer churn, which can be either total (when a customer stops using the e-commerce site altogether) or partial (when a customer reduces their purchases or becomes less profitable), The dynamic churn definition step enables the model to detect the two types of churn. The model uses the Length, Regency, Frequency, and Monitory (LRFM) model combined with the k-means algorithm to define churn status in the first phase. In the second phase of the study, the model uses XGBoost on behavioral and interaction data to predict customer churn status. The results of this study showed that the proposed model achieves an accuracy rate of 98% for the algorithm that detects both partial and total churn, while the accuracy for the partial churn algorithm is 98% and the accuracy for the total churn algorithm is 99%.