Ranking procedures by pairwise comparison using random sets and the imprecise Dirichlet model (Q864784)

The aim of this paper is to investigate methods for ranking of alternatives or objects, relying on the random set theory and the imprecise Dirichlet model. The author considers and provide solutions to the following three problems: (1) Ranking procedures are studied using the comparison of belief and plausibility measures that correspond to each sequence of ranking indices, and an algorithm for computing the belief and plausibility measures for every ranking is developed. (2) The second problem considered is to develop an efficient algorithm for computing the belief and plausibility measures, whose complexity weakly depends on the number of independent sources of data. (3) The third solved problem is to develop a method for computing the cautions, belief and plausibility measures taking into account the small number of judgements. Using the imprecise Dirichlet model and an extension of belief and plausibility measures, the proposed method is shown to overcome some difficulties concerning the conflicting or contradictory sources of data. The main contributions of the paper are considered to be the following ones: the proposed methods allow (a) to correct some incorrect judgements, (b) to use the simplest type of pairwise comparisons, (c) to take into account the possible independence of sources of data, (d) to obtain a tractable computational complexity that weakly depends on the number of data sources, (e) to give cautious decisions when the number of expert judgements is rather small, and (f) to use and compare interval-valued measures and different decision procedures (pessimistic, optimistic or their combination) for choosing the best ranking. Various numerical examples illustrate the proposed algorithms and methods.