A technique to compensate for disparate sources in evidence combination

Abstract

Any decision procedure that is applied to a given problem involves both the gathering of relevant information from different sources and, when possible, accounting for the reliability of each source or group of sources. Due to insufficient knowledge, it may not be possible to isolate an unreliable or poor source of information. However, if a set of sources provides information to a decision algorithm, the effect of these sources as a group could be analyzed in terms of the relative coherence of the sources within it. A group whose sources are in agreement regarding the same decision is more reliable than one in which the individual sources supported different decisions. A measure of the unreliability of a group is its 'disparity', which could compensate for disagreement within a group. Theory and results to support this group unreliability are presented.<>