Three-stage churn management framework based on DCN with asymmetric loss

Abstract

Customer retention is always a hot topic in customer relationship management, and machine learning is widely used in this area. However, existing researches focus on increasing the accuracy of predicting customer churn rather than provide a method of making practical strategies to retain customer in business. Such being the case, this paper proposes a three-stage churn management framework, integrating churn prediction, customer segmentation and strategy making effectively. Two open datasets are collected and applied our framework respectively. In customer prediction phase, a Deep and Cross model taking cross features into consideration is built to learn more implicit information. During assessment, 10-fold cross validation is used, and our model outperforms existing mainstream models. In customer segmentation phase, customers are segmented into different groups using the output of our model and the importance scores of cross features are calculated as well. In strategy making phase, possible strategies for business are made based on the results of the previous phases. Then we propose a fast feature deduction approach to verify the efficiency of strategies, which offers instructive information for decision-makers.