Quantification Classification Algorithm of Multiple Sources of Evidence

Abstract

Although Dempster–Shafer (D–S) evidence theory and its reasoning mechanism can deal with imprecise and uncertain information by combining cumulative evidences for changing prior opinions of new evidences, there is a deficiency in applying classical D–S evidence theory combination rule when conflict evidence appear — conflict evidence causes counter-intuitive results. To address this issue, alternative combination rules have been proposed for resolving the appeared conflicts of evidence. An underlying assumption is that conflict evidences exist, which, however, is not always true. Moreover, it has been verified that conflict factors may not be accurate to characterize the degree of conflict. Instead, the Jousselme distance has been regarded as a quantification criterion for the degree of conflict because of its promising properties. To avoid the counter-intuitive results, multiple sources of evidence should be classified first. This paper proposes a novel algorithm to quantify the classification of multiple sources of evidence based on a core vector method, and the algorithm is further verified by two examples. This study also explores the relationship between complementary information and conflicting evidence and discusses the stochastic interpretation of basic probability assignment functions.