Pattern recognition with evidential knowledge

Many real situations may be seen as classifications problems if the: inform ation used to relate items with classes is a probability measure on P(X x C ) (set of all subsets of X x C). It is well known that a probability measur e on P(X x C) induces two evidence measures on X x C. We will formulate a c lassification model where it is assumed classes are fuzzy sets and the avai lable information is evidential. Two bad classification loss functions will be established for this purpose, by using Dempster's upper and lower integ rals. (C) 1999 John Wiley & Sons, Inc.