Traceable uncertainty

Many applications will benefit greatly when newly encountered situations can be identified as such at run-time. This can be achieved by using the uncertainty arising during the information fusion process. High uncertainty may indicate input data of low quality, but also that the encountered situation is difficult to identify as one of the known situations. In the latter case, the user should be informed about the nature of the uncertainty and about how much evidence supports each of the possible matches. The user can than contribute with context information or expert knowledge, or allocate more resources to clarify the situation. An important precondition for this is that the uncertainty can be traced through the fusion process. Therefore, before deciding on an uncertainty representation, the ability to trace the uncertainty using the representation should be evaluated. In this paper, we provide a method for traceable uncertainty based on evidence theory that, by using an established uncertainty measure, keeps track of increased/decreased uncertainty for evidence combination. Our initial evaluation of the method shows that it is insensitive to noise in the input data and computationally feasible.