A new evidence reliability coefficient for conflict data fusion and its application in classification
Shuaihong Wu, Yongchuan Tang · 2021 IEEE International Conference on Systems, Man, and Cybernetics (SMC) · 2021
Information fusion-based classification is a key issue in many practical applications of automation. Dempster-Shafer evidence theory is a tool to model and fuse uncertain information. However, if there is a high contradiction between two bodies of evidence, the classical Dempster combination rule often gives the fusion result which violates the intuitive results. To address this issue, this paper proposes a single factor belief function with a new evidence reliability coefficient. The original evidence is preprocessed based on the reliability coefficient to get new evidence and to generate a reasonable basic probability assignment (BPA) function. After that, the combined results of the evidence are expressed in the binary form of reliability coefficient and BPA. The merit of the proposed method is that it takes advantage of the characteristics of the evidence itself to deal with the uncertain data and avoids the problem that the classical Dempster combination rule is not applicable for high conflict evidence. Finally, two UCI data sets are used to verify the rationality and effectiveness of the new method.