A Trajectory-based Deep Sequential Method for Customer Churn Prediction

Abstract

Customer churn prediction is a pivotal issue in business marketing. Many researches have been pursuing more efficient features and techniques for it. Rapid growth of mobile Internet devices has generated large amounts of customer trajectory data, which contains abundant customer behavior patterns and contributes to many business actions. In this paper we propose a trajectory-based deep sequential method called TR-LSTM for customer churn prediction to mining the customer behavior pattern behind trajectory data. The method extracts three types of trajectory-based features and applied the long short-term memory neural network (LSTM) to conduct sequential modeling. Experimental results on real-world customer trajectory data sets demonstrate that the proposed TR-LSTM obtains better performance than all baseline methods. Our method provides a new tool of churn prediction for both academics and practitioners.