Detection of defective sources with belief functions

Francois Delmotte, David Gacquer.

This paper studies the fusion of several sources with belief functions. Different operators have been defined but they have problems with conflicting data: rules are either very imprecise, or very sensitive. Discounting factors enable to weight the influences of sources, and solve some problems, but we have to estimate correctly these factors. We propose to estimate them from the conflicts between the sources and from past knowledge about the qualities of sources. With the assumption that conflicts come from defective sources, an algorithm is proposed to detect such sources and to lower conflicts.

PDF full paper