Applying Reinforcement Learning for Customer Churn Prediction. (August 2020)

• Record Type: Journal Article
• Authors: Panjasuchat, M; Limpiyakorn, Y
• Is Part Of: Journal of physics. Volume 1619(2020)
• Publication Date: 2020-08
• Subjects: Physics -- Congresses; 530.5
• DOI: 10.1088/1742-6596/1619/1/012016
• Languages: English
• ISSNs: 1742-6588

Abstract

Abstract: Customer churn prediction is one of the biggest challenges for companies nowadays. Since the loss of customers directly affects the organization's reputation, financial and growth plans. Customer behaviours may change due to any uncontrollable factors or unexpected circumstances, resulting in changing patterns of data. This may aggravate the prediction performance of the classifiers generated from supervised learning technique considered as Passive learning. This paper has thus proposed applying the technique of reinforcement learning for customer churn prediction. The model of Deep Q Network (DQN) is implemented and adapted for learning on the selected customer churn dataset used for classification tasks. We simulate a different distribution dataset with shuffle sampling to embrace pattern changes. The performance of the selected classifiers, compared to DQN, has been evaluated with four measures: accuracy, precision, recall, and F1. The results showed that as Active learner, DQN outperformed the selected classifiers, XGBoost, Random forest, and kNN, when data have been enlarged and evolving with emerging new patterns.