A parameterized family of belief functions correction mechanisms

David Mercier, Thierry Denoeux, Marie-Hélène Masson.

To correct the information, represented by a belief function, provided by a source, different tools can be used such as discounting, de-discounting, extended discounting. In this paper, the links between these operations are explored. A new interpretation of these schemes, as well as a parameterized family of transformations encompassing all previous schemes is introduced and justified. A postal application illustrates the benefits obtained by the use of one of these new correction mechanisms.

PDF full paper