An improvement for combination rule in evidence theory

Evidence theory is an effective tool to make decision from ambiguity, which has been widely used in target recognition, decision making, optimization problem. To reduce its impact on combination results, the conflicting evidence should be assigned to a smaller weight than others when being combined. However, due to the phenomenon of similarity collision, the weight for conflicting evidence probably cannot be reduced effectively in present combination rules for similarity is the main criterion. In this paper, based on the analysis and illustration of similarity collision, a new combination rule is proposed, in which, the impact of similarity collision on evidence weights are reduced obviously by introducing the Basic Probability Assignment sorting before the final combination. In the experiment part, two sets of experiments are designed to show the superiority of the proposed method by comparing the size of each Basic Probability Assignment belonging to the correct decision and the F-Score of classification under the dataset Iris.