Fuzzy Measures and Integrals
Linett Monta~noGuzman
UniversidadCatoli aBoliviana
DepartamentodeCien iasExa tas
Co habamba, Bolivia
e-mail:lmontanou b ba.edu.bo
Abstra t
Inthispaper,wewillstudytheaggregation problemwithintera ting riteria,
and we will introdu e the on epts of fuzzymeasures and integrals. Thefuzzy
integrals are powerful aggregation operators whi h are able to take into
onsi-deration the intera tionamong riteriaand the fuzzymeasures anbe usedfor
modelingtheimportan eofallsubsetsof riteria.
Wewill on entratein\theChoquetintegral",thisfuzzyintegralisaweighted
aggregationoperator,whi htakesintoa ountnotonlytheimportan eofthe
ri-teria, butalsotheimportan eofall subsetsofthem. Theinformationaboutthe
importan eandtheintera tionbetween riteriaisusedintheaggregationpro ess
ofthepartialevaluations. Thispaperanalysesome hara teristi sandproperties
ofthisoperator.
keywords: Multiple riteriaanalysis;Fuzzymeasures;aggregationoperator.
1 Introdu tion
Multi riteriaDe ision Aid(MCDA)is an importantbran h of OperationalResear h.
MCDA dis iplinedevelopsand implements de ision support toolsand methodologies
to dealwith omplex de isionproblems that involvemultiple riteria,goals or
obje -tivesof on i tingnature. ThetoolsandmethodologiesproviedbyMCDAarenotjust
mathemati almodelsaggregating riteria,pointsofvieworattributesbutfurthermore
theyarede isionsupportoriented. MostMulti riteriaDe isionAid(MCDA)methods
needsomewhere in their pro edure afundamental operation: \the aggregation". The
most ommonaggregationtoolusedtoday,isstilltheweightedarithmeti mean. This
operatorhasane essary onditionfortherepresentationofpreferen es: \the
indepen-den e". But,inmanysituations,however,the riteria onsideredarenotindependent,
theyintera t.
In fa t,in ade isionproblem there are links between riteria. Forinstan e, there
ofrelation should be taken into a ountin aggregationmethods and espe ially in the
pro ess ofdetermining the weights. Thus, thefuzzy measures and integralstakeinto
a ountthisinformation.
In the following se tions, we will introdu e someimportant notionsof thetheory
offuzzy measuresand integration. Formoredetails andproofsaboutthistheory, the
readeris referredtothefollowingpapersandbooks[1, 2,3,6,7,8,10℄.
2 Intera tion among riteria
In ade ision problem the evaluation of thealternativesunder various points of view
( riteria) often showsintera tions. These links express the orrelation or the
depen-den e with respe t to the preferen es among riteria. This does not mean that the
orrespondingpointsofviewareredundantandshouldbeeliminated. Ontheopposite,
thismeans, that theattributesthat areused tore e t thesepointsofviewarelinked
by logi alorfa tual interdependen iesand this information should beused. We will
analyse two types of intera tion among riteria: the orrelation and the preferential
dependen e.
2.1 Correlation
Theword orrelationisusedin everydaylifetodenotesomeformofasso iation.
Thestatisti aldenition ofthenotionof orrelationisthefollowing:
Denition2.1 The orrelation measures the inter-relationship between two variables
XandY. Theoutputofthismeasurementis alledthe\ orrelation oeÆ ient",denoted
by r. Itissometimes alled Pearson's orrelation oeÆ ientafter itsoriginatorand is
ameasure oflinearasso iation.
r= n n X i=1 X i Y i ( n X i=1 X i n X i=1 Y i )℄ v u u t [n n X i=1 X 2 i ( n X i=1 X i ) 2 ℄ v u u t [n n X i=1 Y 2 i ( n X i=1 Y i ) 2 ℄ (1)
The orrelation oeÆ ient is measured on a s ale that varies from 1 through 0
to +1. Complete orrelation between two variables is expressed by either +1 or 1.
When one variable in reases as the other in reases the orrelation is positive; when
onede reasesasthe otherin reasesit isnegative. Complete absen eof orrelationis
representedby0.
Instatisti altermsthenotionof orrelationisusedtodenoteanasso iationbetween
twoquantitativevariables. Itisalsoassumedthattheasso iationislinear,thatis,that
onevariablein reasesorde reasesaxedamountforaunitin reaseorde reaseofthe
The orrelationamong riteriaisprobablythebestknownandmostintuitivetypeof
dependen e. Forinstan e,inthesele tionofpersonnelproblem,theageofthe andidate
andthenumberofyearsofprofessionalexperien e, riteriausedin theevaluation,are
likelytobepositively orrelated.
2.2 Preferentialdependen e
Anotherwaytohavedependen eorintera tionamong riteriaiswhentheevaluationof
thealternativesfollowdierentpointsofview,donotrespe tthepropertyofpreferential
independen e.
A simple test, to prove the independen e in the sense of preferen es, onsists in
askingthede isionmakerabouthispreferen es,onpairsofalternatives,thatsharethe
sameprole, forasubset ofattributes;varying the ommonproleshould notreverse
thepreferen eswhenthepointsofviewareindependent.
Let us onsiderer a set of alternatives A = fa;b;:::;mg and a set of riteria
F = fg 1 ;g 2 ;:::;g n
g. Ea h alternative a 2 A is asso iated with a prole g(a) =
(g 1 (a);g 2 (a);:::;g n (a)) 2 IR n where g j
(a) represents the utility of a related to the
riterionj.
Supposethatthepreferen esoverAofthede isionmakerareknownandexpressed
byabinaryrelation.
Denition2.2 (Vin ke1992 [14 ℄)LetF bethefamilyof riteria, S asubsetofF and
S the omplementary subset in F. S ispreferentially independent in F if, given four
a tionsa;b; ;dsu hthat:
8 > > < > > : g j (a)=g j (b) 8j2S g j ( )=g j (d) 8j2S g j (a)=g j ( ) 8j2S g j (b)=g j (d) 8j2S (2) weget aPb, Pd;
whereP isthe global preferen erelationtaking intoa ount allthe riteria.
In other words, S is preferentially independent in F if the preferen es between
a tions, whi h dier only by their evaluations a ording to the riteria in S, do not
depend uponthevaluesyieldedbythe riteriaofS.
Example1 Considerthe sele tion of personnelproblem: A ompany ishiring and
hasashortlistof4 andidates: Anne,Bernard,Celine andDavid. TheHRdepartment
has a list of three riteria: age of the andidate (g
1
), number of years of professional
experien ein arelevant relatedjob (g
2
) andnumber of years of higher edu ation (g
3 ).
Candidate g 1 g 2 g 3 Anne(a) 28 4 6 Bernard(b) 28 6 4 Celine( ) 35 4 6 David (d) 35 6 4
Table1: Sele tionofpersonnelproblem.
The HR department expresses its opinion about the andidates. Evidently, the
preferen es a and b d are immediately suggested; but the other omparisons
are notsoobviousbe ausethe asso iated prolesinterla e. After the analysis of the
problem,theHRdepartmentexpressesthefollowingpreferen es:
andidatea andidateb and andidated andidate
FortheHRdepartment,thereasonisthatastheagein reases,theyearsof
profes-sionalexperien ebe omemoreimportantthantheyearsofhigheredu ation. However,
when the andidate is younger, it is more important to have a high edu ation than
experien e. Inthis asetheprofessional experien eand thenumberof yearsof higher
edu ation are not preferentially independent of the age of the andidate. Thus, the
rsttwo riteriaarenotpreferentiallyindependentofthethird. Therefore,inthis ase,
thereisnoadditivemeasurethat ouldreprodu etheHRdepartmentranking.
A wellknown exampleof dependen e in the sense of preferen es is thepreferen e
forwhitewineorredwinedependingonwhetheryouareeatingsh ormeat.
Inordertotakeintoa ountthebehaviorof riteria: the orrelationanddependen e
in the sense of preferen es and to have a exible representation of these intera tions
between riteria,it isusefulto introdu e the on eptoffuzzy measuresand integrals.
Thus, in thenextse tions weintrodu ethese main on epts, andweshow howfuzzy
measuresprovideameansofrepresentingtherelationshipbetween riteria.
3 Fuzzy measures
Theadditivitypropertyofthe\measure",oneofthemostimportant on eptsin
math-emati s,isoftenin exibleortoorigidtorepresentthemanyfa etsofhumanreasoning.
Therefore,tobeabletoexpresshumansubje tivity,Sugeno[13℄proposedtorepla ethe
additivitypropertybyaweakerone,themonotoni ity,andhe alledthesenon-additive
monotoni measures\fuzzymeasures".
Beforetheintrodu tionoffuzzymeasuresbySugenoin1974(togeneralizeadditive
measures),this on epthasbeenintrodu ed in1953byChoquet[1℄as\ apa ities".
Thus,thefuzzymeasuresarealso allednon-additivemeasures,or apa ities.
Denition3.1 A dis rete \fuzzy measure" on N is a set fun tion : 2 N
! [0;1℄
i) (;)=0; (N)=1
ii) ST N =) (S)(T)
Afuzzymeasure issaidto be:
i) Additiveif(S[T)=(S)+(T)wheneverS\T =;.
ii) Super-additiveif(S[T)(S)+(T)wheneverS\T =;.
iii) Sub-additiveif(S[T)(S)+(T)wheneverS\T =;.
Tounderstandthe meaningof fuzzymeasures in pra ti alappli ations wegive an
exampleproposedbyMurifushiandSugeno[12℄:
Example2 LetN beasetofworkersinaworkshop,andsupposetheyprodu ethe
sameprodu ts. SupposethatagroupofworkersAN worksinthemosteÆ ientway.
Let (A) bethe number of produ tsmade by A in one hour. an be onsideredas a
measureofprodu tivity,and anbenormalizeddividingby(N). Then,thenormalized
fun tion of the set fun tion :2 N
!IR
+
is monotoneand vanishes atthe empty set,
andtherefore itisafuzzymeasure.
Observethatin thissituation, maybenon-additive,sin etwogroupsof workers
AandB together anprodu emore(orless)thaniftheyworkseparately. IfAandB
work separately, then (A[B)= (A)+(B). Nevertheless, generally ifthey work
jointly, theyintera tand theequalitymaynotbekept. Thus,anee tive ooperation
ofmembersofA[B givestheinequality(A[B)>(A)+(B). Ontheotherhand,
the in ompatibility between the groups of workersA and B ould give the opposite
inequality(A[B)<(A)+(B).
Inade isionaidframework,thefuzzymeasure(T)representstheweightof
impor-tan eofthesubsetof riteriaT. Therefore,inadditiontotheusualweightson riteria
takenintoa ountseparately,weightsonallthe ombinationsof riteriawillbedened
aswell.
In the asewhere the fuzzy measure is additive, we an onsider that there is no
intera tionbetween riteriaanditsuÆ estodenenweightsf(1);:::;(n)gtodene
themeasure entirely. Whenthefuzzymeasureissuper-additive,thismeansthatthere
isapositiveintera tionbetween riteria(synergy). Otherwise,ifthefuzzymeasuresis
sub-additiveitmeansthatthereisanegativeintera tionbetween riteria(redundan y).
There are severalrepresentations ofafuzzy measure. A ording to Grabis h[4℄ a
representationofafuzzymeasure ouldbedenedby\anysetfun tionfromwhi h it
ispossibletore overwithoutlossof information". Themostutilizedrepresentations
arethefollowing:
Theusual(measure)representation;thisissimply()itself.
TheMobiusrepresentationorpolynomialrepresentation(w).
4 Fuzzy integral
\Fuzzyintegral"is ageneri termfor integralswith respe t tofuzzy measures. There
aremany kindsof fuzzy integrals: TheintegralsofChoquet,
Sipos,Sugeno,t- onorm
integral,et . Inthispaperwewillmainly on entrateontheChoquetintegral.
Murofushi and Sugenoproposed the so- alled Choquetfuzzy integral, referringto
afun tion dened by Choquetinadierent ontext. TheChoquetintegralisafuzzy
integralbasedonafuzzymeasure that providesalternative omputationals hemefor
aggregatinginformation.
Infa t, thisaggregation operator,in ade isionaid ontext, takesinto a ount, in
theevaluationpro ess,notonlytheimportan eofea h riteria,butalsotheimportan e
ofallsubsets. Theweightsarerepresentedbythe oeÆ ientsofthefuzzymeasure.
Denition4.1 Let, be a fuzzy measure on N. The dis rete Choquet integral of a
fun tion g:N !IR withrespe ttoisdenedby:
C (g (1) ;:::;g (n) )= n X i=1 (g (i) g (i 1) )(A (i) ) (3) where (:)
indi ates that the indi es have been permutated, su h that f0 g
(1) g (2) :::g (n) g; A (i) =f(i);:::;(n)gandg (0) =0.
Figure1: TheChoquetintegral.
Forinstan e, if g 3 g 1 g 2 , then g (1) = g 3 , g (2) = g 1 and g (3) = g 2 , the fuzzy measures(A (1) )=(312);(A (2) )=(12)and(A (3)
)=(2) andthefuzzyintegral
willbe: C (g 1 ;g 2 ;g 3 )=(g 3 g (0) )(3;1;2)+(g 1 g 3 )(1;2)+(g 2 g 3 )(2)
5 Properties of the Choquet integral
Inordertoshowthesuitabilityoffuzzyintegralsinthemulti riteriade isionaid
frame-work,wepresentin thisse tionthemainpropertiesforaggregationofthisoperator.
TheChoquetintegralisamonotoni in reasingfun tion C
dened from [0;1℄ n
in
[0;1℄andhasthefollowingproperties:
i) Boundary onditions: Thispropertyexpressesthebehavioroftheaggregation
operatorin theworstandin thebest ase.
C
(0;:::;0)=0 C
(1;:::;1)=1
ii) Idempoten e: Forallg
j
identi al(=a),thisaggregationoperatorrestitutesthe
ommonvalue(a).
C
(a;:::;a)=a
iii) Continuity : Thispropertymeansthat theChoquetintegralis ontinuouswith
respe ttoea hofitsvariables. Itmeansthat,thisoperatordoesnotpresentany
haoti rea tiontoasmall hangeinthearguments.
iv) Monotoni ity : This operatorisnon-de reasingwithrespe t toea hvariable.
g i g j =)C (g 1 ;:::;g i ;:::g n )C (g 1 ;:::;g j ;:::g n )
v) De omposability : Thispropertymeansthatanysubsetofelementsg2R n
an
berepla edbytheirpartialaggregationwithout hangingtheglobalaggregation.
C (n) (g 1 ;:::;g k ;g k +1 :::;g n )=C (n) (g;:::;g;g k +1 ;:::;g n ) whereg=C (k ) (g i ;:::;g k )
vi) Stability under the samepositive linear transformations: This property
translatesastabilityoftheoperatorfora hangeofmeasurements ale.
C (rg 1 +t;:::;rg n +t)=rC (g 1 ;:::;g n )+t 8r>0;8t2IR :
Fromthis property, we an say that hangingthe s aledoesnot hange the
re-sult. This is an extremely important property in utility theory, given that the
evaluationsofana tiona ordingto ea h riterionaredened withrespe tto a
positivelineartransformation. Theglobalutilityofthea tionhavethustokeep
thisproperty.
Fromboundary onditionsi),idempoten yii)andmonotoni ityiv)properties,the
Lemma 1 The Choquet integrals are omprised between the minimun and the maximunvalue. minfg 1 ;:::;g n gC (g)maxfg 1 ;:::;g n g
Proof: From boundary onditionsi),idempoten yii)and monotoni ityiv)
prop-ertieswehave: Forg=fg 1 ;:::;g n g Let g f
=minfggbethe minimumvalueof g, and g
k
=maxfggbethemaximum
valueofg g f g i 8i2f1;:::;ng g k g i 8i2f1;:::;ng fg f ;:::;g f gfg 1 ;:::;g n gfg k ;:::;g k g
Fromthemonotoni itypropertywehave:
C (g f ;:::;g f )C (g 1 ;:::;g n )C (g k ;:::;g k )
Fromtheidempoten yproperty
g f C (g)g k asg f =min(g)andg k =max(g)then min(g)C (g)max(g)
Fromthislemma,we ansaythatthisaggregationoperatorisa\ ompromise
oper-ator".
Inthisse tionwementionedthepropertiesofthefuzzyintegralsrelatedtothe
rep-resentationofintera tionsbetween riteria. Now,we anillustrate,whatisunderstood
byintera tionsandhowthey anbemodelledbyfuzzyintegral,withasimpleexample.
Example2 Considerthesameexampleofthesele tionofpersonnelproblemshown
in se tion 2.2, with other andidates, A, B and C. The HR department has alist of
three riteria: age of the andidate (g
1
), number of years of professional experien ein
arelevantrelatedjob(g
2
)andnumberofyearsofhigheredu ation(g
3
). Theevaluation
involvingthese andidates isdes ribed inthe Table 2.
The HRdepartmentwantswellequilibrated andidateswithoutweakevaluations.
Thismeans,thattheideal andidateshouldbeyoung,withvariousyearsofprofessional
experien eandwithahigh levelofedu ation.
The normalized matrix 1
and the weight sum results are shown in Table 3. The
rankingofthistableisnotfullysatisfa toryfortheHRdepartment,sin ethe andidate
1
The riteriong1 shouldbe minimized,and g2;g3 maximized. We hoosehere to maximizethe
inversevalues 1
g1
andforthenormalization,wedivideg
i
bythetotalsum,i.eg 0 i = g i P n i g i
Candidate g 1 g 2 g 3 A 28 7 3 B 36 2 7 C 29 5 5
Table 2: Sele tionofpersonnelproblem.
Candidate g 1 g 2 g 3 Globalevaluation (weighted sum) A 0,36 0,50 0,20 0,36 B 0,28 0,14 0,47 0,29 C 0,35 0,36 0,33 0,35 Weight 0,35 0,35 0,3
Table3: Normalizedmatrixandweightedsumresults.
Ahasaweaknessinthenumberofyearsofhigheredu ation(seeTable2),buthasbeen
onsidered better than andidate C, whohas no weak point. The reason is that too
mu h importan eisgivento ageandthenumberofyearsofprofessionalexperien eof
the andidates,whi hareredundant. Usually, theageandtheprofessionalexperien e
haveastronglink. SolvingthisproblemwiththeChoquetintegralandanappropriate
fuzzymeasurewehave:
i) FortheHRdepartment,theageandtheprofessionalexperien earemore
impor-tantthanthenumberofyearsofhigheredu ation.
(fg 1 g)=(fg 2 g)=0:35 (fg 3 g)=0:3
ii) Be ause the riteria g
1 and g
2
are redundant, the weight attributed to the set
fg
1 ;g
2
gshould belessthanthesumoftheweightsofthepairof riteria.
(fg
1 ;g
2
g)=0:45<0:35+0:35
iii) Sin e, the riteria g
1 ;g 3 and g 2 ;g 3
have a positive intera tion, the weight
at-tributed to theset fg
1 ;g 2 gandfg 2 ;g 3
gshould begreaterthanthe sumof
indi-vidualweights. (fg 1 ;g 3 g)=0:8>0:35+0:3 (fg 2 ;g 3 g)=0:8>0:35+0:3 And(fg 1 ;g 2 g 3 g)=1.
Table4showstheresultsapplyingthisfuzzymeasureto the andidates.
TheHR ansee thattheChoquetintegralgivestheexpe tedresults.
This example exposes how easy it is to translatethe requirementsof the de ision
on-Candidate g 1 g 2 g 3 Globalevaluation (Choquet integral) A 0,36 0,50 0,20 0,32 B 0,28 0,14 0,47 0,31 C 0,35 0,36 0,33 0,34
Table 4: Choquetintegralresults.
sidersmore than 3 riteria, the evaluation of the fuzzy measure be omes diÆ ult. Is
forthisreasonthatthenextse tionintrodu ethek-orderadditivefuzzymeasureand
integral.
6 k-order additive fuzzy measure and integral
Inade isionaidproblem involvingn riteria, ifwetakeintoa ounttheintera tions
between riteria, we need to dene the 2 n
oeÆ ients of the fuzzy measure
orre-sponding tothe2 n
subsets ofN. Thede isionmakerisnotableto givesu h amount
ofinformation. Inmost ases,hewill beabletoguesstheimportan eofea h riterion
andofea hpairof riterion.
Toover omethisproblem,Grabis h[5℄proposedtousethe on eptofk-orderfuzzy
measure.
Denition6.1 (Grabis h(97) [5 ℄) A fuzzy measure dened on N is said to be
k-orderadditive if its orresponding pseudo-Boolean fun tion isamultilinearpolynomial
ofdegreek,i.e. itsMobiustransformw(T)=0forall(T)su hthat(T)>k,andthere
existatleastone subsetT of N ofexa tly k elementssu hthat w(T)6=0.
Thus, for the2-additivefuzzy measures, oeÆ ients of the Mobius representation
aregivenby
(i) = w(i); i2N; (4)
(ij) = w(i)+w(j)+w(ij); fi;jg2N (5)
The monotoni ity onstraints and the normalisation of the oeÆ ients of the
2-additivefuzzymeasureareformulatedasfollows:
8 > > > > > > < > > > > > > : (;)=0; X fi;jg2N (ij)+(n 2) X i2N (i)=1; (i)0; 8i;2N X j2S (ij) X j2S (j) (n 2)(i)0; 8i2N;8SNni: (6)
Intermsof theMobiustransformthese onstraintsare: 8 > > > > > > < > > > > > > : w(;)=0; X i2N w(i)+ X fi;jg2N w(ij)=1; w(i)0; 8i;2N w(i)+ X j2S w(ij)0; 8i2N;8SNni: (7)
Ifwe onsiderthe2-ordermodel,theChoquetintegralbe omes:
C (g)= X i2N w(i)g i + X fi;jgN w(ij)(g i ^g j ) 8g2IR n (8)
The2-order aseoftheChoquetintegral,seemstobeinterestinginpra ti al
appli- ations. It permits tomodel intera tionbetween riteriawhileremainingverysimple.
Indeed,onlyn+( n
2
)=(n(n+1))=2 oeÆ ientsarerequiredtodenethefuzzymeasure.
Therearemanyappli ationsbasedonthe2-ordermodel[1, 2,3,6,7,8,9,10℄.
Areviewoftheliterature[3,6,7,11℄,showsthattherearemainlythreeapproa hes
tondthefuzzymeasures.
Identi ation basedonsemanti s
Identi ationbasedonlearningdata
Identi ationbasedonthe ombinationofthesemanti sandlearningdata.
Therstapproa hisbasedonthede isionmaker'sinterpretationofthefuzzy
mea-sure. The se ond one is established on the assignment by the de ision makerof the
numeri als oreforea ha tionandea h riterion,andalsoanumeri alglobals orefor
ea h a tion,from this informationit ispossibletobuild alearningsystem tondthe
fuzzymeasure. Finallythethirdapproa hisbasedonthe ombinationofthesemanti s
andlearningdata,indeedthispro edure ombinessemanti al onsiderationsand
learn-ingdatawhi hintrodu esveryusefulinformation,redu esthe omplexityandprovides
moreeÆ ientalgorithmsto ndthefuzzymeasure.
7 MCDA methods and Fuzzy measures and integrals
Aswehaveshowninthispaper,inmanysituations,the riteria onsideredinanMCDA
evaluation pro essarenotindependent,theyintera t. Inthesesituationsthefuzzy
in-tegralbe omestheappropriateaggregationoperatortodealwiththiskindofproblems.
Forthisreasonisveryinterestingtheintrodu tionofthispowerfulaggregationoperator,
Referen es
[1℄ G.Choquet. Theoryof apa ities. Annalesde l'InstitutFourier,5:131{295,1953.
[2℄ M. Grabis h. Fuzzy integral in multi riteria de ision making. Fuzzy Sets and
Systems,69:279{298,1994.
[3℄ M.Grabis h. The appli ation offuzzy integralsin multi riteriade isionmaking.
EuropeanJournalof Operational Resear h,89:445{456,1995.
[4℄ M. Grabis h. Alternative representations of dis rete fuzzy measure for de ision
making. International Journal of Un ertainty, Fuzines and Knowledge - Based
Systems, 5:587{607,1997.
[5℄ M. Grabis h. k-order additive dis rete fuzzy measure and their representation.
Pattern Re ognition Letters,92:167{189,1997.
[6℄ M.Grabis h,T.Murofushi,andM.Sugeno.FuzzyMeasuresandIntegrals-Theory
andAppli ations. Physi aVerlag,2000.
[7℄ M.Grabis h,H.T.Nguyen,andE.A.Walker.Fundamentalsofun ertainty al uli
withappli ations tofuzzyinferen e. Kluwera ademi publishers,London,1995.
[8℄ M.Grabis h andJ.M.Ni olas. Classi ationbyfuzzy integral-performan eand
test. Fuzzy Sets and Systems, Spe ial Issue on Pattern Re ognition, 65:255{271,
1994.
[9℄ L. MontanoGuzman. FuzzyMeasures and Integrals in the MCDA Sorting
Prob-lemati : Methodology and Appli ation tothe Diagnosis of Firms. TesisDo toral,
UniversiteLibredeBruxelles,2002.
[10℄ J.L.Mari hal.Aggregationoperatorsformulti riteriade isionaid. TesisDo toral,
UniversitedeLiege,1998-1999.
[11℄ J.L. Mari hal and M. Roubens. Determination of weights of intera ting riteria
from a referen eset. European Journal of Operational Resear h,124(3):641{650,
1999.
[12℄ T.MurofushiandM.Sugeno.Aninterpretationoffuzzymeasuresandthe hoquet
integralasan integralwith respe t toa fuzzymeasure. FuzzySets andSystems,
29:201{227,1989.
[13℄ M. Sugenoand T.Murofushi. Choquetintegralasanintegralform for ageneral
lass offuzzymeasures.2nd IFSACongress,1987.