Premium
Stochastic Ranking of Alternatives with Ordered Weighted Averaging: Comparing Network Recovery Strategies
Author(s) -
Rocco Claudio M.,
Barker Kash,
HernándezPerdomo Elvis
Publication year - 2016
Publication title -
systems engineering
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.474
H-Index - 50
eISSN - 1520-6858
pISSN - 1098-1241
DOI - 10.1002/sys.21367
Subject(s) - weighting , randomness , ranking (information retrieval) , decision maker , computer science , probability distribution , rank (graph theory) , percentile , mathematical optimization , cumulative distribution function , extension (predicate logic) , mathematics , point (geometry) , data mining , econometrics , statistics , operations research , machine learning , probability density function , medicine , geometry , combinatorics , radiology , programming language
Comparing system design alternatives can be difficult when the criterion used to evaluate and rank the alternatives exhibits randomness described by a probability distribution. Rather than relying on the expected value to compare alternatives, this work proposes to use a multicriteria decision‐making technique, ordered weighted averaging (OWA), for such a comparison. Here, percentiles of the cumulative distributions describing the evaluation criterion act as the “multiple criteria” in OWA, incorporating information from the entirety of the distribution, not just a lone point estimate. OWA allows for the elicitation of risk preferences by the decision maker, and a weighted OWA extension allows for the elicitation of importance weighting for each of the percentiles (e.g., weighting tail values of the distribution more heavily to account for adverse outcomes). The technique is applied to the comparison of link recovery sequences in a network disruption problem, along with a visualization aid to assist in accounting for uncertainty in importance weighting.