Deep reinforcement learning with the confusion-matrix-based dynamic reward function for customer credit scoring
作者：
Highlights：
• Customer credit scoring is formulated as a finite Markov decision process.
• The confusion-matrix-based dynamic reward function (CMDRF) is constructed.
• CMDRF can accelerate the convergence speed and improve the stability of the Deep Q-network (DQN) model.
• The DQN-CMDRF model is proposed for customer credit scoring.
• The proposed model achieves excellent performance in customer credit scoring.
摘要
•Customer credit scoring is formulated as a finite Markov decision process.•The confusion-matrix-based dynamic reward function (CMDRF) is constructed.•CMDRF can accelerate the convergence speed and improve the stability of the Deep Q-network (DQN) model.•The DQN-CMDRF model is proposed for customer credit scoring.•The proposed model achieves excellent performance in customer credit scoring.
论文关键词：Deep reinforcement learning,Deep Q-network,Dynamic reward function,Confusion matrix,Customer credit scoring
论文评审过程：Received 29 June 2021, Revised 14 November 2021, Accepted 27 March 2022, Available online 4 April 2022, Version of Record 5 April 2022.
论文官网地址：https://doi.org/10.1016/j.eswa.2022.117013
