Transfer Graph in the Context of Evidence Theory

Tianxiang Zhan, Qianzhou Zhou, Yuanpeng He, Yong Deng

Abstract: Evidence theory is a framework for uncertainty reasoning that has made contributions in the field of artificial intelligence. There are two key issues in Evidence Theory: inability to find all inverse mass distributions from Pignistic Probability Transformation (PPT), and PPT may not be suitable for all sample distributions in the decision-making process. To address these key issues, Transfer Graph (TG) on Evidence Theory are proposed. Firstly, TG first constrains Mass Distributions with the same PPT results into a fixed dimensional space, solving the problem of being unable to find them. Secondly, by optimizing the Cross Entropy, Mass Transfer Learning (MTL) is proposed to achieve the optimal decision-making process. The experimental results indicate that MTL can optimize decision-making process to a certain extent, improve the accuracy of classification experiments.