Geoscience Reference
In-Depth Information
criteria has to be measured. Let us take the example of the selection of a plot of land
for the construction of a new housing project. The set of alternatives (A) corresponds
tothedifferentplotsoflandofferedforsale: plot
1
,...,plot
n
.Thesetoffeatures(A/F)
determines the quality of such a plot in a property context: 1-low noise pollution, 2-
unfloodable zone, 3-affordable prices, 4-immediate availability, etc. The evaluation
oftheperformanceofeachoftheavailableplotswithrespecttothesecriteriagenerally
results in the attribution of scores.
Different approaches to criterion aggregation are presented and illustrated
in [PAL 10a]. The evaluation of the criteria can be based on a type of Boolean
calculation or proportional quantitative calculation. Such approaches, for which all
the criteria are of equal importance, allow us an alternative less satisfying criteria to
be ranked equally well or better because of a higher score of the satisfied
criteria [YAG 88, MAL 03]. Consequently, preferences can be associated with the
criteria. These allow us to promote certain criteria and influence the mode of
compensation [MAL 03, BÜY 10]. Let us note that the obligation to satisfy a given
criterion is not taken into account.
In parallel, there are numerous studies which mainly look to attenuate the
compensation created by the definition of preferences and the quantification of
criteria-alternative couples. The ordered weighted averaging (OWA)
method [YAG 88] allows us to specify the level of compensation between criteria.
For example, the OWA “orand” operator proposes sorting the criteria by order of
decreasing scores and defining the weight for each rank instead of each criterion, the
total weight not exceeding one (an example of OWA is described in [PAL 10a]). This
method enables us, by weighting, to parametrize an approach by total compensation,
medium compensation or without compensation. Pereira et al. [DA 09, DA 12]
propose a method allowing the use of four criteria to evaluate the relevance of a
document: aboutness, coverage, appropriateness and reliability. Besides associating
a weight with each of these criteria, they also associate an order of priority with
them. For example, when the aboutness is the criterion of priority number 1 and it is
not satisfied, the other criteria are not examined. In a more general way, whatever the
criteria, the order in which they are expressed corresponds to the level of importance
the user wishes to assign to them; Pereira et al. [DA 09, DA 12] emphasize the
particularly intuitive characteristics of this approach.
All these approaches consider that the different criteria are independent. The
weighted arithmetic average proposed in most of these studies does not allow us to
take into account the possible relations existing between criteria. Several published
studies propose the use of methods such as the Choquet integral
[BÜY 10, LAB 03, HAM 09]. For example, Büyüközkan and Ruan [BÜY 10]
explain that, in the framework of risk evaluation in software development, some risks
(criteria) are linked. Instead of associating a weight with a criterion, the Choquet
integral allows us to assign weights to pairs of criteria. As explained by Labreuche
Search WWH ::
Custom Search