A new method to determine basic probability assignment using core samples

摘要

The Dempster-Shafer theory of evidence (D S theory) has been widely used in many information fusion systems. However, the determination of basic probability assignment (BPA) remains an open problem which can considerably influence final results. In this paper, a new method to determine BPA using core samples is proposed. Unlike most of existing methods that determining BPA in a heuristic way, the proposed method is data-driven. It uses training data to generate core samples for each attribute model. Then, helpful core samples in generating BPAs are selected. Calculation of the relevance ratio based on convex hulls is integrated into the core sample selection as a new feature of the proposed method. BPAs are assigned based on the distance between the test data and the selected core samples. Finally, BPAs are combined to get a final BPA using the Dempster's combination rule. In this paper, compound hypotheses are taken into consideration. BPA generated by the proposed method can be combined with some other sources of information to reduce the uncertainty. Empirical trials on benchmark database shows the efficiency of the proposed method.