Evidence generalization-based discounting method: assigning unreliable information to partial ignorance

• Main Authors: Hu, Qiying, Zhou, Qianli, Li, Zhen, Deng, Yong, Cheong, Kang Hao
• Other Authors: College of Computing and Data Science
• Format: Article
• Language: English
• Published: 2024
• Subjects: Computer and Information Science; Discounting; Generalization of belief function
• Online Access: https://hdl.handle.net/10356/180845
• Institution: Nanyang Technological University

Conflict management is an important topic when dealing with unreliable sources information fusion in Dempster–Shafer theory. Discounting unreliable bodies of evidence has proven to be effective to decrease conflict. Based on the generalization of belief functions, a new generalization-based discounting method is proposed. When resolving conflicts with the same degree, our method can realize less information loss in comparison with other discounting methods. By simulating the process of resolving conflicts of randomly generated bodies of evidence, using entropy measurements and binary conflict as evaluation index, we show our method’s rationality and superiority. Finally, the hyperparameters of the conflict metrics are generated and generalization-based discounting is applied to classify real-world datasets. The improved classification performance further illustrates the usefulness of the method.