Learning the parameters of a multiple criteria sorting method from large sets of assignment examples

ELECTRE TRI is a sorting method used in mul
tiple criteria décision analysis. It assigns each alternative, de-
scribed by a performance vector, to a category selected in a
set of pre-defined ordered catégories. Consécutive catégories
are separated by a profile. In a simplified version proposed
and studied by Bouyssou and Marchant and called MR-Sort,
a majority rule is used for assigning the alternatives to caté
gories. Each alternative a is assigned to the lowest category
for which a is at least as good as the lower profile delimiting
this category for a majority of weighted criteria. In this pa-
per, a new algorithm is proposed for learning the parameters
of this model on the basis of assignment examples. In contrast
with previous work ([8]), the présent algorithm is designed to
deal with large learning sets. Expérimental results are pre-
sented, which assess the algorithm performances with respect
to issues like model retrieval, computational efficiency and tolerance for error.