An improved jousselme evidence distance

Abstract

Evidence distance can be used in uncertainty simulation result validation, especially when the simulation model has epistemic uncertainty. Jousselme evidence distance (JED) and a modified Jousselme evidence distance (MJED) are introduced and analyzed first. Then the paper points out problems that they have when they measure the difference between two bodies of evidence: JED will measure two bodies of general evidence unreasonably; The distance value of MJED will be too large sometimes. An improved Jousselme evidence distance (IJED) is proposed to solve these problems. IJED can measure the dissimilarly between category evidence and general evidence as JED and MJED. And the distance value of IJED is within reasonable range when it measures two bodies of general evidence. Furthermore, IJED can measure the difference better when probability distribution parameters of the variable for the body of evidence change. Two numerical examples are illustrated to compare the three distances.