Learning valued preference structures for solving classification problems

摘要: This paper introduces a new approach to classification which combines pairwise decomposition techniques with ideas and tools from fuzzy preference modeling. More specifically, our first decomposes polychotomous problem involving m classes into an ensemble of binary problems, one for each ordered pair classes. The corresponding classifiers are trained on the relevant subsets (transformed) original training data. In phase, query is submitted every learner. output classifier interpreted as degree in comparison second class. By combining outputs all classifiers, thus obtains relation taken point departure final decision. way, effectively reduced decision making based relation. Corresponding techniques, have been investigated quite intensively field set theory, hence become amenable task classification. particular, by decomposing strict preference, indifference, incomparability relation, this allows quantify different types uncertainty thereby supports sophisticated postprocessing strategies.