A multi-granularity distance with its application for decision making

Abstract

In decision-making systems, conflict management is an important concept that represents the degree of dissimilarity between bodies of evidence, ultimately enhancing decision-making performance. Jousselme's distance, as the most commonly employed one so far, is used to measure the distance between basic belief assignments (BBAs) in Dempster-Shafer (D-S) evidence theory. However, the Jousselme's distance has limitations, which can also be demonstrated in other methods theoretically. To address this issue, a BBA refinement method and a novel multi-granularity distance are proposed in this paper. Moreover, the methods are verified to be effective in the problems that Jousselme's distance cannot satisfy. Additionally, a hypothetical physical model is employed to verify the practicability of the proposed methods with multiple granularity. Furthermore, based on the proposed multiple granularity distance, a novel decision-making algorithm is designed. The results validate that the proposed decision-making method is beneficially applicable to classification scenarios and different real-world data.