An adaptive evidence combination method for decision analysis under uncertainty

Due to the imperfection of devices and the individuation of human cognition, the process of data fusion often involves uncertainty. Dempster–Shafer theory defines the basic probability assignments of possible hypotheses and is effective in combining uncertain information from multiple sources. However, the existing evidence combination methods lack the flexibility to achieve compensation between conflicting pieces of evidence. This study aims to propose an adaptive evidence combination method that takes into account the personalized compensation requirements of decision makers in solving problems of conflicting evidence. To achieve this, an adjustment coefficient is added to the basic probability assignment of each hypothesis to control the compensation degrees between conflicting pieces of evidence in a flexible manner. The parameters of information reliability and importance are further incorporated into the model. The algebraic properties of the proposed evidence combination method are described. In addition, we conduct two case studies, one on vehicle recognition based on multiple sensors and one on purchasing decisions based on online reviews. Through the sensitivity analysis of the adjustment coefficient and the comparative analysis with other evidence combination methods, the advantages of the proposed method in dealing with high levels of conflicting evidence are verified.