Enhanced mass Jensen–Shannon divergence for information fusion. (15th December 2022)

• Record Type: Journal Article
• Authors: Pan, Lipeng; Gao, Xiaozhuan; Deng, Yong; Cheong, Kang Hao
• Is Part Of: Expert systems with applications. Volume 209(2022)
• Publication Date: 2022-12-15
• Subjects: Mass function -- Conflict -- Enhanced mass Jensen–Shannon divergence -- Information fusion; Expert systems (Computer science) -- Periodicals; Systèmes experts (Informatique) -- Périodiques; Electronic journals; 006.33
• DOI: 10.1016/j.eswa.2022.118065
• Languages: English
• ISSNs: 0957-4174

Abstract

Abstract: Conflict issue has been a topic of immense interest in evidence theory because the current methods still do not accurately reflect the conflict degree between evidence bodies. Thus, this paper defines mass Jensen–Shannon divergence (MJSD) to reflect the conflict degree. MJSD considers both the external difference and the external difference of the mass function. Furthermore, this paper proposes an enhanced mass Jensen–Shannon divergence (EMJSD). EMJSD has all the advantages of MJSD. It is a distance-like measure, which satisfies some axiom of distance measure, such as non-negativity and symmetry. Compared with the current methods, EMJSD fully measures the internal and external difference between the mass functions. Therefore, EJMSD overcomes the drawbacks of the classical methods when measuring differences between mass functions. Some numerical examples are used to explain its properties and advantages. Additionally, this paper presents a new method to fuse multi-source information. A numerical example shows that this fusion method is more useful for decision-making under the framework of multi-source information fusion. We then construct a new generation method of mass function in the data-driven environment. The experimental results show that when compared with other method, the new generation method more accurately reflects the support degree of proposition. Finally, this paper applies the generated mass functions to the new fusion method and other methods. The results indicate that the new fusion method has better recognition performance in practical applications.