An approximate approach for ranking fuzzy numbers based on left and right dominance (Q5948812): Difference between revisions

The paper starts with a detailed critical review of the literature on comparing fuzzy numbers. The authors favor Yager's F3 index for its intuitive nature and generality. They argue that this index requires the knowledge of all the membership functions involved in the comparison, which might be impractical or might result in high computational cost. Others of the surveyed methods do not require the knowledge of the membership functions but they depend heavily on the choice of their parameters, which is also inconvenient. The authors define a fuzzy number as a normal convex fuzzy set on a bounded subset of the real line \(\mathbb{R}\). The fuzzy number is specified by its left (increasing) and right (decreasing) membership functions. The authors consider a series of \(n\) alpha-cuts of the fuzzy numbers to determine the dominant fuzzy number. Each alpha-cut is a (crisp) bounded subset of \(\mathbb{R}\), whose limits are used to calculate the left and right dominance. The total dominance index is proposed to be a convex combination of the left and right dominance. Equal offset between the adjacent alpha-levels is suggested. When the number of the alpha-cuts, \(n\), tends to infinity, the proposed comparison index is shown to coincide with F3 and other indices from the literature. Two examples are given to illustrate the ability of the proposed method to rank discrete fuzzy numbers and mixed fuzzy and crisp numbers. The efficiency of the proposed method consists in the less amount of comparisons when multiple fuzzy numbers are being ranked. This is demonstrated by an example taken from the literature. In a comparative experiment, two groups of examples with 13 comparison tasks in each, are processed by 32 different ranking methods. Discrepancies in the results are identified and commented upon, pointing at some inadequacies of some ranking methods found in the literature. The authors argue that there is no need for a large number of alpha levels for their method to work well (suggested is \(n=6\)).