Chapter 7

UTA METHODS
Yannis Siskos
University of Piraeus
Department of Informatics
80 Karaoli & Dimitriou Str.
18534 Piraeus !reece
ysis"os#unipi.$r
Evangelos Grigoroudis, Nikolaos F. Matsatsinis
%ec&nical University of 'rete
Decision Support Systems (a)oratory
University 'ampus* Kounoupi+iana
,3100 '&ania !reece
-van$elis*ni"os.#er$asya.tuc.$r
Abstract !"A #ethods re$er to the philosoph% o$ assessing a set o$ value or utilit%
$unctions, assu#ing the a&io#atic basis o$ MA!" and adopting the pre$erence
disaggregation principle. !"A #ethodolog% uses linear progra##ing
techni'ues in order to opti#all% in$er additive value(utilit% $unctions, so that
these $unctions are as consistent as possible )ith the global decision*#aker+s
pre$erences ,in$erence principle-. "he #ain ob.ective o$ this chapter is to
anal%ticall% present the !"A #ethod and its variants and to su##ari/e the
progress #ade in this $ield. "he historical background and the philosoph% o$
the aggregation*disaggregation approach are $irstl% given in this chapter. "he
detailed presentation o$ the basic !"A algorith# is presented, including the
discussion on the stabilit% and sensitivit% anal%ses. Several variants o$ the
!"A #ethod, )hich incorporate di$$erent $or#s o$ opti#alit% criteria used in
the 01 $or#ulation, are also discussed. "he i#ple#entation o$ the !"A
#ethods is illustrated b% a general overvie) o$ !"A*based 2SSs, as )ell as
real*)orld decision*#aking applications. Finall%, several potential $uture
research develop#ents o$ the !"A #ethodologies )ithin the conte&t o$
MC2A are discussed through this chapter.
3 '&apter ,
4e% )ords !"A #ethods, 1re$erence 2isaggregation, 5rdinal 6egression, Additive
!tilit%, Multicriteria Anal%sis
1. INTRODUCTION
1.1 General philosophy
7n decision*#aking involving #ultiple criteria, the basic proble# stated
b% anal%sts and decision*#akers concerns the )a% that the $inal decision
should be #ade. 7n #an% cases, ho)ever, this proble# is posed in the
opposite )a% assu#ing that the decision is given, ho) is it possible to $ind
the rational basis $or the decision being #ade8 5r e'uivalentl%, ho) is it
possible to assess the decision9#aker+s pre$erence #odel leading to e&actl%
the sa#e decision as the actual one or at least the #ost :si#ilar; decision8
"he philosoph% o$ pre$erence disaggregation in #ulticriteria anal%sis is to
assess(in$er pre$erence #odels $ro# given pre$erential structures and to
address decision*aiding activities through operational #odels )ithin the
a$ore#entioned $ra#e)ork.
!nder the ter# :#ulticriteria anal%sis; t)o basic approaches have been
developed involving
<. a set o$ #ethods or #odels enabling the aggregation o$ #ultiple
evaluation criteria to choose one or #ore actions $ro# a set / and
3. an activit% o$ decision*aid to a )ell*de$ined decision*#aker
,individual, organi/ation, etc.-
7n both cases, the set / o$ potential actions ,or ob.ects, alternatives,
decisions- is anal%/ed in ter#s o$ #ultiple criteria in order to #odel all the
possible i#pacts, conse'uences or attributes related to the set /.
6o% ,<=>?- outlines a general #odeling #ethodolog% o$ decision*#aking
proble#s, )hich includes $our #odeling steps starting )ith the de$inition o$
the set / and $inishing )ith the activit% o$ decision*aid, as $ollo)s
9 (evel 1 5b.ect o$ the decision, including the de$inition o$ the set o$
potential actions / and the deter#ination o$ a proble#atic on /.
9 (evel 0 Modeling o$ a consistent $a#il% o$ criteria assu#ing that
these criteria are non*decreasing value $unctions, e&haustive and non*
redundant.
9 (evel 3 2evelop#ent o$ a global pre$erence #odel, to aggregate the
#arginal pre$erences on the criteria.
9 (evel 4 2ecision*aid or decision support, based on the results o$
level @ and the proble#atic o$ level <.
,. U%/ 1et&o+s @
7n level <, 6o% ,<=>?- distinguishes $our re$erence proble# state#ents,
each o$ )hich does not necessaril% preclude the others. "hese proble#atics
can be e#plo%ed separatel%, or in a co#ple#entar% )a%, in all phases o$ the
decision*#aking process. "he $our proble#atics are the $ollo)ing
9 Pro)lematic 2 Choosing one action $ro# / ,choice-.
9 Pro)lematic 3 Sorting the actions into pre*de$ined and pre$erence*
ordered categories ,sorting-.
9 Pro)lematic 4 6anking the actions $ro# the best one to the )orst one
,ranking-.
9 Pro)lematic 5 2escribing the actions in ter#s o$ their per$or#ances
on the criteria ,description-.
7n level 3, the #odeling process #ust conclude on a consistent $a#il% o$
criteria { }
< 3
, , ,
n
$ $ $ K . Each criterion is a non*decreasing real valued
$unction de$ined on /, as $ollo)s
A
A
B , C ( , -
i i i
$ / $ $ a $ a → ⊂ → ∈ ¡ ¡
)here B
A
A
,
i i
$ $ C is the criterion evaluation scale,
A i
$ and
A
i
$ are the )orst
and the best level o$ the i*th criterion respectivel%, , -
i
$ a is the evaluation or
per$or#ance o$ action
a
on the i*th criterion and
, - a g
is the vector o$
per$or#ances o$ action
a
on the n criteria.
Fro# the above de$initions the $ollo)ing pre$erential situations can be
deter#ined
( )
( )
, - , - is pre$erred to
, - , - is indi$$erent to
i i
i i
$ a $ ) a ) a )
$ a $ ) a ) a )
¹ > ⇔
¹
'
· ⇔
¹
¹
f
:
So, having a )eak*order pre$erence structure on a set o$ actions, the
proble# is to ad.ust additive value or utilit% $unctions based on #ultiple
criteria, in such a )a% that the resulting structure )ould be as consistent as
possible )ith the initial structure. "his principle underlies the
disaggregation*aggregation approach presented in the ne&t section.
"his chapter is devoted to !"A #ethods, )hich are regression based
approaches that have been developed as an alternative to #ultiattribute
utilit% theor% ,MA!"-. !"A #ethods not onl% adopt the aggregation*
disaggregation principles, but the% #a% also be considered as the #ain
initiative and the #ost representative e&a#ple o$ pre$erence disaggregation
theor%. Another, #ore recent e&a#ple o$ the pre$erence disaggregation
theor% is the do#inance*based rough set approach ,26SA- leading to
decision rule pre$erence #odel via inductive learning ,see chapter <? o$ this
book-.
D '&apter ,
1.2 The disaggregaion!aggregaion paradig"
7n the traditional aggregation paradig#, the criteria aggregation #odel is
kno)n a priori, )hile the global pre$erence is unkno)n. 5n the contrar%, the
philosoph% o$ disaggregation involves the in$erence o$ pre$erence #odels
$ro# given global pre$erences ,6i$ure ,71-.
C67"E67A
G05EA0
16EFE6ENCE
Aggregation #odel
Aggregation #odel 8
6i$ure ,71. "he aggregation and disaggregation paradig#s in MC2A ,Fac'uet*0agrG/e and
Siskos, 3HH<-
"he disaggregation*aggregation approach ,Fac'uet*0agrG/e and Siskos,
<=>3I 3HH<I Siskos, <=>HI Siskos and Yannacopoulos, <=>?I Siskos et al.,
<==@- ai#s at anal%/ing the behavior and the cognitive st%le o$ the 2ecision
Maker ,2M-. Special iterative interactive procedures are used, )here the
co#ponents o$ the proble# and the 2M+s global .udg#ent polic% are
anal%/ed and then the% are aggregated into a value s%ste# ,6i$ure ,70-. "he
goal o$ this approach is to aid the 2M to i#prove his(her kno)ledge about
the decision situation and his(her )a% o$ pre$erring that entails a consistent
decision to be achieved.
7n order to use global pre$erence given data, Fac'uet*0agrG/e and Siskos
,3HH<- note that the clari$ication o$ the 2M+s global pre$erence necessitates
the use o$ a set o$ re$erence actions
8
/ . !suall%, this set could be
<. a set o$ past decision alternatives ,
8
/ past actions-,
3. a subset o$ decision actions, especiall% )hen / is large ,
8
/ / ⊂ -,
@. a set o$ $ictitious actions, consisting o$ per$or#ances on the criteria,
)hich can be easil% .udged b% the decision*#aker to per$or# global
co#parisons ,
8
/ $ictitious actions-.
,. U%/ 1et&o+s ?
7n each o$ the above cases, the 2M is asked to e&ternali/e and(or con$ir#
his(her global pre$erences on the set
8
/ taking into account the
per$or#ances o$ the re$erence actions on all criteria.
Consistenc% o$ the
pre$erence #odel and
2M+s .udg#ent polic%
1re$erence #odel
construction
2ecision data * 2M +s
global .udg#ent polic%
Criteria #odeling
2ecision
1roble#
2M+s pre$erences
1roble#
Aggregation o$
2M+s pre$erences
on the criteria
Jalue or utilit%
s%ste#
2ecision
9a: %&e value system approac&
2M+s pre$erences
1roble#
Aggregation o$
2M+s pre$erences
and construction o$
outranking relations
E&ploitation o$ the
constructed
outranking relations
Aid the 2M to #ake
a :good; decision
9): %&e outran"in$ relation approac&
9c: %&e +isa$$re$ation7a$$re$ation approac&
Multiob.ective
#athe#atical
progra##ing
$or#ulation
2M+s satis$action
level and(or utilit%
#odel
Multiob.ective
opti#i/ation
techni'ues
2ecision
9+: %&e multio);ective optimi<ation approac&
7nteractive procedure
6i$ure ,70. "he disaggregation*aggregation approach vs. other MC2A approaches ,Siskos
and Sp%ridakos, <===-
K '&apter ,
1.# Hisori$al %a$&gro'nd
"he histor% o$ the disaggregation principle in #ultidi#ensional(
#ulticriteria anal%ses begins )ith the use o$ goal progra##ing techni'ues, a
special $or# o$ linear progra##ing structure, in assessing(in$erring
pre$erence(aggregation #odels or in developing linear or non9linear
#ultidi#ensional regression anal%ses ,Siskos, <=>@-.
Charnes et al. ,<=??- proposed a linear #odel o$ opti#al esti#ation o$
e&ecutive co#pensation b% anal%/ing or disaggregating pair)ise
co#parisons and given #easures ,salaries-I the #odel )as esti#ated so that
it could be as consistent as possible )ith the data $ro# the goal progra##ing
point o$ vie).
4arst ,<=?>- #ini#i/ed the su# o$ absolute deviations via goal
progra##ing in linear regression )ith one variable, )hile Lagner ,<=?=-
generali/es the 4arst+s #odel in the #ultiple regression case. 0ater 4elle%
,<=?>- proposed a si#ilar #odel to #ini#i/e the "cheb%che$$+s criterion in
linear regression.
Srinivasan and Shoker ,<=7@- outlined the 5626EG ordinal regression
#odel to assess a linear value $unction b% disaggregating pair)ise
.udg#ents. Freed and Glover ,<=><- proposed goal progra##ing #odels to
in$er the )eights o$ linear value $unctions in the $ra#e o$ discri#inant
anal%sis ,proble#atic M-.
"he research on handling ordinal criteria began )ith the studies o$ Young
et al. ,<=7K-, and Fac'uet*0agrG/e and Siskos ,<=7>-. "he latter research
re$ers to the presentation o$ the !"A #ethod in the :Cahiers du
0AMSA2E; series and indicates the actual initiation o$ the develop#ent o$
disaggregation #ethods. Eoth research tea#s $aced the sa#e proble# to
in$er additive value $unctions b% disaggregating a ranking o$ re$erence
alternatives. Young et al. ,<=7K- proposed alternating least s'uares
techni'ues, )ithout ensuring, ho)ever, that the additive value $unction is
opti#all% consistent )ith the given ranking. 7n the case o$ the !"A #ethod,
opti#alit% is ensured through linear progra##ing techni'ues.
2. THE UTA METHOD
2.1 (rin$iples and noaion
"he !"A ,!"ilitNs Additives- #ethod proposed b% Fac'uet*0agrG/e and
Siskos ,<=>3- ai#s at in$erring one or #ore additive value $unctions $ro# a
given ranking on a re$erence set
8
/ . "he #ethod uses special linear
,. U%/ 1et&o+s 7
progra##ing techni'ues to assess these $unctions so that the ranking,s-
obtained through these $unctions on
8
/ is ,are- as consistent as possible
)ith the given one.
"he criteria aggregation #odel in !"A is assu#ed to be an additive
value $unction o$ the $ollo)ing $or# ,Fac'uet*0agrG/e and Siskos, <=>3-
<
, - , -
n
i i i
i
u p u $
·
·

g
sub.ect to nor#ali/ation constraints
<
A
A
<
, - H, , - < <, 3, ,
n
i
i
i i i i
p
u $ u $ i n
·
¹
·
¹
'
¹
· · ∀ · …
¹

)here
i
u , <, 3, , i n · K are non*decreasing real valued $unctions, na#ed
#arginal value or utilit% $unctions, )hich are nor#ali/ed bet)een H and <,
and
i
p is the )eight o$
i
u ,6i$ure ,73-.
H
, -
i i
u $
A i
$
A
i
$
<
criterion $
i
6i$ure ,73. "he nor#ali/ed #arginal value $unction
Eoth the #arginal and the global value $unctions have the #onotonicit%
propert% o$ the true criterion. For instance, in the case o$ the global value
$unction the $ollo)ing properties hold
B , -C B , -C ,pre$erence-
B , -C B , -C ,indi$$erence-
u a u ) a )
u a u ) a )
> ⇔ ¹
'
· ⇔
¹
g g
g g
f
:
"he !"A #ethod in$ers an un)eighted $or# o$ the additive value
$unction, e'uivalent to the $or# de$ined $ro# relations ,7*@- and ,7*D-, as
$ollo)s
> '&apter ,
<
, - , -
n
i i
i
u u $
·
·

g
sub.ect to nor#ali/ation constraints
A
<
A
, - <
, - H <, 3, ,
n
i i
i
i i
u $
u $ i n
·
¹
·
¹
'
¹
· ∀ · …
¹

5$ course, the e&istence o$ such a pre$erence #odel assu#es the
pre$erential independence o$ the criteria $or the 2M ,4eene% and 6ai$$a,
<=7K-, )hile other conditions $or additivit% have been proposed b% Fishburn
,<=KK, <=K7-. "his assu#ption does not pose signi$icant proble#s in a
posteriori anal%ses such as disaggregation anal%ses.
2.2 De)elop"en o* he UTA "ehod
5n the basis o$ the additive #odel ,7*K-*,7*7- and taking into account the
pre$erence conditions ,7*?-, the value o$ each alternative
8
a / ∈ #a% be
)ritten as
<
B , -C B , -C , -
n
i i 8
i
u a u $ a a a / σ
·
′ · + ∀ ∈

g
)here
, - a σ
is a potential error relative to B , -C u a ′ g .
Moreover, in order to esti#ate the corresponding #arginal value
$unctions in a piece)ise linear $or#, Fac'uet*0agrG/e and Siskos ,<=>3-
propose the use o$ linear interpolation. For each criterion, the interval
A
A
B , C
i i
$ $ is cut into , <-
i
α − e'ual intervals, and thus the end points
;
i
$ are
given b% the $or#ula
( )
A
A A
<
<, 3, ,
<
;
i i i i i
i
;
$ $ $ $ ; α
α

· + − ∀ ·

K
"he #arginal value o$ an action
a
is appro&i#ated b% a linear
interpolation, and thus, $or
<
, - B , C
; ;
i i i
$ a $ $
+

<
<
, -
B , -C , - , - , -
;
; ; ; i i
i i i i i i i i ; ;
i i
$ a $
u $ a u $ u $ u $
$ $
+
+

1 · + −
¸ ]

"he set o$ re$erence actions
< 3
O , , P
8 m
/ a a a · K is also :rearranged; in
such a )a% that
<
a is the head o$ the ranking and
m
a its tail. Since the
ranking has the $or# o$ a )eak order 8 , $or each pair o$ consecutive actions
<
, , -
" "
a a
+
it holds either
< " "
a a
+
f ,pre$erence- or
< " "
a a
+
: ,indi$$erence-.
"hus, i$
< <
, , - B, , -C B, , -C
" " " "
a a u a u a
+ +
′ ′ ∆ · − g g
then one o$ the $ollo)ing holds
,. U%/ 1et&o+s =
< <
< <
, , - i$$
, , - H i$$
" " " "
" " " "
a a a a
a a a a
δ
+ +
+ +
∆ ≥ ¹
'
∆ ·
¹
f
:
)here δ is a s#all positive nu#ber so as to discri#inate signi$icantl% t)o
successive e'uivalence classes o$ 8 .
"aking into account the h%pothesis on #onotonicit% o$ pre$erences, the
#arginal values , -
i i
u $ #ust satis$% the set o$ the $ollo)ing constraints
<
, - , - <, 3, , <, <, 3, ,
; ;
i i i i i i
u $ u $ s ; i n α
+
− ≥ ∀ · − · K K
)ith H
i
s ≥ being indi$$erence thresholds de$ined on each criterion
i
$ .
Fac'uet*0agrG/e and Siskos ,<=>3- urge that it is not necessar% to use these
thresholds in the !"A #odel , H-
i
s · , but the% can be use$ul in order to
avoid pheno#ena such as
<
, - , -
; ;
i i i i
u $ u $
+
· )hen
< ; ;
i i
$ $
+
f .
"he #arginal value $unctions are $inall% esti#ated b% #eans o$ the
$ollo)ing linear progra# ,01- )ith ,7*K-, ,7*7-, ,7*<3-, ,7*<@- as constraints
and )ith an ob.ective $unction depending on the
, - a σ
and indicating the
a#ount o$ total deviation
< <
< <
<
A
<
B#inC , -
sub.ect to
, , - i$
, , - H i$
, - , - H and
, - <
8
a /
" " " "
" " " "
; ;
i i i i
n
i i
i
6 a
a a a a
a a a a
u $ u $ i ;
u $
u
σ
δ

+ +
+ +
+
·
·
∆ ≥
∆ ·
− ≥ ∀
·


f
:
A
, - H, , - H, , - H , and
;
i i i i 8
$ u $ a a / i ; σ
¹
¹
¹
¹
¹
¹
¹
'
¹
¹
¹
¹
¹
¹
· ≥ ≥ ∀ ∈ ∀
¹
"he stabilit% anal%sis o$ the results provided b% 01 ,7*<D- is considered
as a post*opti#alit% anal%sis proble#. As Fac'uet*0agrG/e and Siskos ,<=>3-
note, i$ the opti#u#
A
H 6 · , the pol%hedron o$ ad#issible solutions $or
, -
i i
u $ is not e#pt% and #an% value $unctions lead to a per$ect
representation o$ the )eak order 8 . Even )hen the opti#al value
A
6
is
strictl% positive, other solutions, less good $or 6 , can i#prove other
satis$actor% criteria, like 4endall+s
τ
.
As sho)n in 6i$ure ,74, the post*opti#al solutions space is de$ined b%
the pol%hedron
A A
, -
all the constraints o$ 01 ,7*<D-
6 6 " 6 ¹ ≤ +
'
¹
)here
A
, - " 6 is a positive threshold )hich is a s#all proportion o$
A
6
.
"he algorith#s )hich could be used to e&plore the pol%hedron ,7*<?- are
branch and bound #ethods, like reverse si#ple& #ethod ,Jan de 1anne,
<H '&apter ,
<=7?-, or techni'ues dealing )ith the notion o$ the lab%rinth in graph theor%,
such as "arr%+s #ethod ,Charnes and Cooper, <=K<-, or the #ethod o$
Manas and Nedo#a ,<=K>-. Fac'uet*0agrG/e and Siskos ,<=>3-, in the
original !"A #ethod, propose the partial e&ploration o$ pol%hedron ,7*<?-
b% solving the $ollo)ing 01s
A A
B#inC , - B#a&C , -
in and in <, 3, ,
pol%hedron ,7*<?- pol%hedron ,7*<?-
i i i i
u $ u $
i n
¹ ¹
¹ ¹
∀ · …
' '
¹ ¹
¹ ¹
6=6
>
?"96
>
:
6=6
>
pol%hedron o$ constraints ,7*<D-
@
0
@
1
6i$ure ,74. 1ost*opti#alit% anal%sis ,Fac'uet*0agrG/e and Siskos, <=>3-
"he average o$ the previous 01s #a% be considered as the $inal solution
o$ the proble#. 7n case o$ instabilit%, a large variation o$ the provided
solutions appears, and this average solution is less representative. 7n an%
case, the solutions o$ the above 01s give the internal variation o$ the )eight
o$ all criteria
i
$ , and conse'uentl% give an idea o$ the i#portance o$ these
criteria in the 2M+s pre$erence s%ste#.
2.# The UTASTAR algorih"
"he !"AS"A6 #ethod proposed b% Siskos and Yannacopoulos ,<=>?-
is an i#proved version o$ the original !"A #odel presented in the previous
section. 7n the original version o$ !"A ,Fac'uet*0agrG/e and Siskos, <=>3-,
$or each packed action
8
a / ∈ , a single error
, - a σ
is introduced to be
#ini#i/ed. "his error $unction is not su$$icient to #ini#i/e co#pletel% the
dispersion o$ points all around the #onotone curve o$ 6i$ure ,75. "he
,. U%/ 1et&o+s <<
proble# is posed b% points situated on the right o$ the curve, $ro# )hich it
)ould be suitable to subtract an a#ount o$ value(utilit% and not increase the
values(utilities o$ the others.
overesti#ation
error A

underesti#ation
error A
Q
H <
<
3
@
.
.
. 6
a
n
k
i
n
g
Global value
6i$ure ,75. 5rdinal regression curve ,ranking versus global value-
7n !"AS"A6 #ethod, Siskos and Yannacopoulos ,<=>?- introduced a
double positive error $unction, so that $or#ula ,7*>- beco#es
<
B , -C B , -C , - , -
n
i i 8
i
u a u $ a a a a / σ σ
+ −
·
′ · − + ∀ ∈

g
)here σ
+
and σ

are the overesti#ation and the underesti#ation error
respectivel%.
Moreover, another i#portant #odi$ication concerns the #onotonicit%
constraints o$ the criteria, )hich are taken into account through the
trans$or#ations o$ the variables
<
, - , - H <, 3, , and <, 3, , <
; ;
i; i i i i i
B u $ u $ i n ; α
+
· − ≥ ∀ · · − K K
and thus, the #onotonicit% conditions ,7*<@- can be replaced b% the non*
negative constraints $or the variables i;
B
,$or H
i
s · -.
Conse'uentl%, the !"AS"A6 algorith# #a% be su##ari/ed in the
$ollo)ing steps
Step 1C
E&press the global value o$ re$erence actions B , -C
"
u a g , <, 3, , " m · K ,
$irst in ter#s o$ #arginal values , -
i i
u $ , and then in ter#s o$ variables i;
B

according to the $or#ula ,7*<7-, b% #eans o$ the $ollo)ing e&pressions
<3 '&apter ,
<
<
<
, - H <, 3, ,
, - <, 3, , and 3, @, , <
i i
;
;
i i it i
t
u $ i n
u $ B i n ; α

·
¹ · ∀ ·
¹
'
· ∀ · · −
¹
¹

K
K K
Step 0C
7ntroduce t)o error $unctions σ
+
and σ

on
8
/ b% )riting $or each pair
o$ consecutive actions in the ranking the anal%tic e&pressions
<
< < <
, , - B , -C , - , -
B , -C , - , -
" " " " "
" " "
a a u a a a
u a a a
σ σ
σ σ
+ −
+
+ −
+ + +
∆ · − +
− + −
g
g
Step 3C
Solve the linear progra#
<
< <
< <
<
< <
B#inC B , - , -C
sub.ect to
, , - i$

, , - H i$
<
H, , - H, , - H , and
i
m
" "
"
" " " "
" " " "
n
i;
i ;
i; " "
< a a
a a a a
"
a a a a
B
B a a i ; "
α
σ σ
δ
σ σ
+ −
·
+ +
+ +

· ·
+ −
¹
¹
· +
¹
¹
¹
¹
∆ ≥ ¹
¹

' ;
∆ ·
¹ ¹
¹
¹
·
¹
¹
≥ ≥ ≥ ∀
¹
¹

∑∑
f
:
)ith δ being a s#all positive nu#ber.
Step 4C
"est the e&istence o$ #ultiple or near opti#al solutions o$ the linear
progra# ,7*3<- ,stabilit% anal%sis-I in case o$ non uni'ueness, $ind the #ean
additive value $unction o$ those ,near- opti#al solutions )hich #a&i#i/e the
ob.ective $unctions
<
A
<
, - <, 3, ,
i
i i i;
;
u $ B i n
α −
·
· ∀ ·

K
on the pol%hedron o$ the constraints o$ the 01 ,7*3<- bounded b% the ne)
constraint
A
<
, - , -
m
" "
"
a a < σ σ ε
+ −
·
1 + ≤ +
¸ ]

)here
A
<
is the opti#al value o$ the 01 in step @ and
ε
a ver% s#all
positive nu#ber.
A co#parison anal%sis bet)een !"A and !"AS"A6 algorith#s is
presented b% Siskos and Yannacopoulos ,<=>?- through a variet% o$
e&peri#ental data. !"AS"A6 #ethod has provided better results
concerning a nu#ber o$ co#parison indicators, like
,. U%/ 1et&o+s <@
<. "he nu#ber o$ the necessar% si#ple& iterations $or arriving at the
opti#al solution.
3. "he 4endall+s
τ
bet)een the initial )eak order and the one produced
b% the esti#ated #odel.
@. "he #ini#i/ed criterion
<
,su# o$ errors- taken as the indicator o$
dispersion o$ the observations.
2.+ A n'"eri$al e,a"ple
"he i#ple#entation o$ the !"AS"A6 algorith# is illustrated b% a
practical e&a#ple presented b% Siskos and Yannacopoulos ,<=>?-. "he
proble# concerns a 2M )ho )ishes to anal%/e the choice o$ transportation
#eans during the peak hours ,ho#e*)ork place-. Suppose that the 2M is
interested onl% in the $ollo)ing three criteria
<. price ,in #onetar% units-,
3. ti#e o$ .ourne% ,in #inutes-, and
@. co#$ort ,possibilit% to have a seat-.
"he evaluation in ter#s o$ the previous criteria is presented in %a)le ,71,
)here it should be noted that the $ollo)ing 'ualitative scale has been used
$or the co#$ort criterion H ,no chance o$ seating-, Q ,little chance o$ seating-
QQ ,great chance o$ $inding a seating place-, and QQQ ,seat assured-. Also,
the last colu#n o$ %a)le ,71 sho)s the 2M+s ranking )ith respect to the $ive
alternative #eans o$ transportation.
%a)le ,71. Criteria values and ranking o$ the 2M
Means o$
transportation
1rice ,#u- "i#e ,#in- Co#$ort 6anking o$ the
2M
6E6 @ <H Q <
ME"65 ,<- D 3H QQ 3
ME"65 ,3- 3 3H H 3
E!S K DH H @
"AR7 @H @H QQQ D
"he $irst step o$ !"AS"A6, as presented in the previous section,
consists o$ #aking e&plicit the utilities o$ the $ive alternatives. For this
reason the $ollo)ing scales have been chosen
[ ]
A
<A <
, @H,<K, 3 $ $ 1 ·
¸ ]
[ ]
A
3A 3
, DH, @H, 3H,<H $ $ 1 ·
¸ ]
[ ]
A
@A @
, H, , , $ $ 1 · + ++ + + +
¸ ]
!sing linear interpolation $or the criterion
<
$ according to $or#ula ,7*
<H-, the value o$ each alternative #a% be )ritten as
<D '&apter ,
< < 3 @
B ,6E6-C H.H7 ,<K- H.=@ ,3- ,<H- , - u u u u u · + + + + g
< < 3 @
B ,ME"65<-C H.<D ,<K- H.>K ,3- ,3H- , - u u u u u · + + + ++ g
< 3 @
< 3
B ,ME"653-C ,3- ,3H- ,H-
,3- ,3H-
u u u u
u u
· + +
· +
g
< < 3 @
< <
B ,E!S-C H.3= ,<K- H.7< ,3- ,DH- ,H-
H.3= ,<K- H.7< ,3-
u u u u u
u u
· + + +
· +
g
< 3 @
3 @
B ,"AR7-C ,@H- ,@H- , -
,@H- , -
u u u u
u u
· + + + + +
· + + + +
g
)here the $ollo)ing nor#ali/ation conditions $or the #arginal value
$unctions have been used
< 3 @
,@H- ,DH- ,H- H u u u · · · .
Also, according to $or#ula ,7*<=-, the global value o$ the alternatives
#a% be e&pressed in ter#s o$ the variables i;
B

<< <3 3< 33 3@ @<
<< <3 3< 33 @< @3
<< <3 3< 33
<< <3
3< @< @3 @@
B ,6E6-C H.=@
B ,ME"65<-C H.>K
B ,ME"653-C
B ,E!S-C H.7<
B ,"AR7-C
u B B B B B B
u B B B B B B
u B B B B
u B B
u B B B B
· + + + + +
· + + + + +
· + + +
· +
· + + +
g
g
g
g
g
According to the second step o$ the !"AS"A6 algorith#, the $ollo)ing
e&pressions are )ritten, $or each pair o$ consecutive actions in the ranking
<3 3@ @3
6E6 6E6 ME"65< ME"65<
<3 @< @3
ME"65< ME"65< ME"653 ME"653
<3 3< 33
ME"653 ME"653 E!S E!S
,6E6, ME"65<- H.H7

,ME"65<, ME"653- H.<D

,ME"653, E!S- H.3=

,E!S, "AR7
B B B
B B B
B B B
σ σ σ σ
σ σ σ σ
σ σ σ σ



+ + −
+ + −
+ + −
∆ · + −
− + + −
∆ · − + +
− + + −
∆ · + +
− + + −

<< <3 3< @< @3 @@
E!S E!S "AR7 "AR7
- H.7<

B B B B B B
σ σ σ σ

+ + −
· + − − − −
− + + −
Eased on the a$ore#entioned e&pression, a linear progra# according to
,7*3<- is $or#ulated, )ith H.H? δ · ,see %a)le ,70-. An opti#al solution is
<<
H.? B · ,
3<
H.H? B · ,
3@
H.H? B · ,
@@
H.D B · )ith
A
B#inC H < < · · . "his
solution corresponds to the #arginal value $unctions presented in %a)le ,73
and produces a ranking )hich is consistent )ith the 2M+s initial )eak order.
%a)le ,70. 7nitial linear progra##ing $or#ulation
B<< B<3 B3< B33 B3@ B@< B@3 B@@ variables A
Q
and A
*
6SS
H H.H7 H H < H *< H *< < < *< H H H H H H T H.H?
,. U%/ 1et&o+s <?
B<< B<3 B3< B33 B3@ B@< B@3 B@@ variables A
Q
and A
*
6SS
H *H.<D H H H < < H H H *< < < *< H H H H U H
H H.3= < < H H H H H H H H *< < < *< H H T H.H?
< H.7< *< H H *< *< *< H H H H H H *< < < *< T H.H?
< < < < < < < < H H H H H H H H H H U <
< < < < < < < < H H H H H H H H H H <
7t should be e#phasi/ed that this solution is not uni'ue. "hrough post*
opti#alit% anal%sis ,step D-, the !"AS"A6 algorith# searches $or #ultiple
opti#al solutions, or #ore generall%, $or near opti#al solutions
corresponding to error values bet)een
A
<
and
A
< ε + . For this reason, the
error ob.ective should be trans$or#ed to a constraint o$ the t%pe ,7*3@-.
%a)le ,73. Marginal value $unctions ,initial solution-
1rice "i#e Co#$ort
u<,@H- U H.HHH u3,DH- U H.HHH u@,H- U H.HHH
u<,<K- U H.?HH u3,@H- U H.H?H u@,Q- U H.HHH
u<,3- U H.?HH u3,3H- U H.H?H u@,QQ- U H.HHH
u3,<H- U H.<HH u@,QQQ- U H.DHH
7n the presented nu#erical e&a#ple, the initial linear progra# has
#ultiple opti#al solutions, since
A
H < · . "hus, in the post*opti#alit%
anal%sis step, the algorith# searches $or #ore characteristic solutions, )hich
#a&i#i/e the e&pressions ,7*33-, i.e. the )eights o$ each criterion.
Further#ore, in this particular case )e have
A
H , - , - H
" "
< a a " σ σ
+ −
· ⇔ · · ∀
so the error variables #a% be e&cluded $ro# the linear progra#s o$ the post*
opti#alit% anal%sis. %a)le ,74 presents the $or#ulation o$ the linear
progra#s that have to be solved during this step.
%a)le ,74. 0inear progra##ing $or#ulation ,post*opti#alit% anal%sis-
B<< B<3 B3< B33 B3@ B@< B@3 B@@ 6SS
H H.H7 H H < H *< H T H.H?
H *H.<D H H H < < H U H
H H.3= < < H H H H T H.H?
< H.7< *< H H *< *< *< T H.H?
< < < < < < < < U <
< < H H H H H H B#a&C u<,$<
A
-
H H < < < H H H B#a&C u3,$3
A
-
H H H H H < < < B#a&C u@,$@
A
-
"he solutions obtained during post*opti#alit% anal%sis are presented in
%a)le ,75. "he average o$ these three solutions is also calculated in the last
ro) o$ %a)le ,75. "his centroid is taken as a uni'ue utilit% $unction,
<K '&apter ,
provided that it is considered as a #ore representative solution o$ this
particular proble#.
%a)le ,75. 1ost*opti#alit% anal%sis and $inal solution
B<< B<3 B3< B33 B3@ B@< B@3 B@@
B#a&C u<,$<
A
- H.7K3? H.<7? H H H.H@7? H.H3? H H
B#a&C u3,$3
A
- H.H? H H H.H? H.= H H H
B#a&C u@,$@
A
- H.@?K3 H.<7? H H H.H@7? H.H3? H H.DHK@
Average H.@>=K H.<<K7 H H.H<K7 H.@3?H H.H<K7 H H.<@?D
"his $inal solution corresponds to the #arginal value $unctions presented
in %a)le ,7D. Also, the utilities $or each alternative are calculated as $ollo)s
B ,6E6-C H.>?K
B ,ME"65<-C H.?3@
B ,ME"653-C H.?3@
B ,E!S-C H.D7@
B ,"AR7-C H.<?3
u
u
u
u
u
·
·
·
·
·
g
g
g
g
g
)here it is obvious that these values are consistent )ith the 2M+s )eak
order.
%a)le ,7D. Marginal value $unctions ,$inal solution-
1rice "i#e Co#$ort
u<,@H- U H.HHH u3,DH- U H.HHH u@,H- U H.HHH
u<,<K- U H.@=H u3,@H- U H.HHH u@,Q- U H.H<7
u<,3- U H.?HK u3,3H- U H.H<7 u@,QQ- U H.H<7
u3,<H- U H.@D3 u@,QQQ- U H.<?3
"hese #arginal utilities #a% be nor#ali/ed b% dividing ever% value
, -
;
i i
u $ b%
A
, -
i i
u $ . 7n this case the additive utilit% can be )ritten as
< < 3 3 @ @
, - H.?HK , - H.@D3 , - H.<?3 , - u u $ u $ u $ · + + g
)here the nor#ali/ed #arginal value $unctions are presented in 6i$ure ,.D.
#. -ARIANTS O. THE UTA METHOD
#.1 Alernai)e opi"aliy $rieria
Several variants o$ the !"A #ethod have been developed, incorporating
di$$erent $or#s o$ global pre$erence or di$$erent $or#s o$ opti#alit% criteria
used in the linear progra##ing $or#ulation.
,. U%/ 1et&o+s <7
An e&tension o$ the !"A #ethods, )here
B , -C u a g
is in$erred $ro#
pair)ise co#parisons is proposed b% Fac'uet*0agrG/e and Siskos ,<=>3-.
"his sub.ective pre$erence obtained b% pair)ise .udg#ents is #ost o$ten not
transitive, and thus, the #odi$ied #odel #a% be )ritten as in the $ollo)ing
01
<H
<
3H @H DH
H
H.HD=
"i#e ,#in-
u
3
,
$
3
-
H
<
Q QQ QQQ
H
H.<<H H.<<H
Co#$ort
u
@
,
$
@
-
3
<
<K @H
H
H.77H
1rice ,#onetar% units-
u
<
,
$
<
-
6i$ure ,7D. Nor#ali/ed #arginal value $unctions
<> '&apter ,
[ ] [ ] { }
[ ] [ ] { } ( )
, , - , , -
<
<
<
B#inC
sub.ect to
, - , - H i$
, - , - H i$
, - , - ,
,
a) a) a) )a
a ) a ) a ) a )
n
i i i i a)
i
n
i i i i a) )a
i
; ;
i i i i i
i i
6 < <
u $ a u $ ) < a )
u $ a u $ ) < < a ) ) a
u $ u $ s i ;
u $
λ λ
·
·
+
· +
− + ≥
− + − · ⇒
− ≥ ∀
∑ ∑


f :
f
: :
A
<
A
- <
, - H, , - H, H , and , , -
n
i
;
i i i i a)
u $ u $ < i ; a ) 8
·
¹
¹
¹
¹
¹
¹
¹
¹
¹
'
¹
¹
¹
¹
·
¹
¹
¹ · ≥ ≥ ∀ ∈
¹
¹

a)
λ being a non negative )eight re$lecting a degree o$ con$idence in the
.udg#ent bet)een
a
and ) .
An alternative opti#alit% criterion )ould be to #ini#i/e the nu#ber o$
violated pairs o$ an order 8 provided b% the 2M in ranking 8′ given b% the
#odel, )hich is e'uivalent to #a&i#i/e 4endall+s
τ
bet)een the t)o
rankings. "his e&tension is given b% the #i&ed integer 01 ,7*3?-, )here
H
a)
γ · i$
B , -C B , -C u a u ) δ − ≥ g g
$or a pair
, , - a ) 8 ∈
and the .udg#ent is
respected, other)ise <
a)
γ · and the .udg#ent is violated. "hus, the
ob.ective $unction in this 01 represents the nu#ber o$ violated pairs in the
overall pre$erence aggregated b%
, - u g
.
[ ] [ ] { }
, , -
<
<
A
<
A
B#inC B#a&C , , -
sub.ect to
, - , - , , -
, - , - ,
, - <
, - H, , - H
a)
a ) 8
n
i i i i a)
i
; ;
i i i i i
n
i i
i
;
i i i i
6 8 8
u $ a u $ ) 1 a ) 8
u $ u $ s i ;
u $
u $ u $
γ τ
γ δ

·
+
·
′ · ⇔
− + ⋅ ≥ ∀ ∈
− ≥ ∀
·
· ≥



,
H or < , , -
a)
i ;
a ) 8 γ
¹
¹
¹
¹
¹
¹
¹
¹
'
¹
¹
¹
¹

¹
¹
· ∀ ∈
¹
¹
)here 1 is a large nu#ber. Eeuthe and Scanella ,3HH<- propose to handle
separatel% the pre$erence and indi$$erence .udg#ents, and #odi$% the
previous 01 using the constraints
,. U%/ 1et&o+s <=
[ ] [ ] { }
[ ] [ ] { }
[ ] [ ] { }
<
<
<
, - , - i$
, - , - H
i$
, - , - H
n
i i i i a)
i
n
i i i i a)
i
n
i i i i )a
i
u $ a u $ ) 1 a )
u $ a u $ ) 1
a )
u $ a u $ ) 1
γ δ
γ
γ
·
·
·
¹
− + ⋅ ≥
¹
¹
¹
¹ ¹
− + ⋅ ≥
'
¹
¹
¹
;
¹
¹
− + ⋅ ≥
¹
¹
¹ ¹ ¹



f
:
"he assu#ption o$ #onotonicit% o$ pre$erences, in the conte&t o$
separable value $unctions, #eans that the #arginal values are #onotonic
$unctions o$ the criteria. "his assu#ption, although )idel% used, is
so#eti#es not applicable to real*)orld situations. 5ne )a% to deal )ith non*
#onotonic pre$erences is to divide the range o$ the criteria into intervals, so
that the pre$erences are #onotonic in each interval, and then treat each
interval separatel% ,4eene% and 6ai$$a, <=7K-. 7n the sa#e spirit, 2espotis
and Vopounidis ,<==@- present a variation o$ the !"AS"A6 #ethod $or the
assess#ent o$ non*#onotonic #arginal value $unctions. 7n this #odel, the
range o$ each criterion is divided into t)o intervals ,see also 6i$ure ,7,-
H
, -
i i
u $
<
A i i
$ $ ·
3
i
$
<
i
p
i
$
+
... i
p
i i
+ $ ·
...
A
i i
p E
i i
$ $
+
·
i
$
6i$ure ,7,. A non*#onotonic partial utilit% $unction ,2espotis and Vopounidis, <==@-
{ }
{ }
< < 3
A
< 3 A
, , ,
, , ,
i
i i i i
p
i i i i i i
p p p E
i i i i i i
! $ $ $ $ +
! + $ $ $ $
+ +
¹
· · ·
¹
'
· · ·
¹
¹
K
K
)here
i
+ is the #ost desirable value o$
i
$ , and the para#eters
i
p and
i
E
are deter#ined according to the dispersion o$ the input dataI o$ course it
holds that
i i i
p E α + · . 7n this approach, the #ain #odi$ication concerns the
assess#ent o$ the decision variables i;
B
o$ the 01 ,7*3<-. Sence, $or#ula
,7*<=- beco#es
3H '&apter ,
<
<
< <
<
i$ <
, -
i$
i
i
;
it i
t
;
i i
p ;
it it i i
t t p
B ; p
u $
B B p ; α

·
− −
· ·
¹
< ≤
¹
¹
·
'
¹
− < ≤
¹
¹

∑ ∑
)hile the conditions
<
, - H
i i
u $ · re#ain.
Another e&tension o$ the !"A #ethods re$ers to the intensit% o$ the
2M+s pre$erences, si#ilar to the conte&t proposed b% Srinivasan and Socker
,<=7@-. 7n this case, a series o$ constraints #a% be added during the 01
$or#ulation. For e&a#ple, i$ the pre$erence o$ alternative
a
over alternative
) is stronger than the pre$erence o$ ) over
c
, then the $ollo)ing condition
#a% be )ritten
[ ] [ ] B , -C B , -C B , -C B , -C u a u ) u ) u c ϕ ′ ′ ′ ′ − − − ≥ g g g g
)here
H ϕ >
is a #easure o$ pre$erence intensit% and , - u′ g is given b%
$or#ula ,7*>-. "hus, using $or#ula ,7*<<-, the $ollo)ing constraint should
be added in 01 ,7*<D-
, , - , , - a ) ) c ϕ ∆ − ∆ ≥
7n general, i$ the 2M )ishes to e&pand these pre$erences to the )hole set
o$ alternatives, a #ini#u# nu#ber o$ 3 m− constraints o$ t%pe ,7*@H- is
re'uired.
2espotis and Vopounidis ,<==@- consider the case )here the 2M ranks
the alternatives using an e&plicit overall inde& I . "hus, $or#ula ,7*<3- #a%
be replaced b% the $ollo)ing condition
< <
, , - <, 3, , <
" " " "
a a I I " m
+ +
∆ · − ∀ · − K
Eesides the succession o$ the alternatives in the pre$erence ranking, these
constraints state that the di$$erence o$ global value o$ an% successive
alternatives in the ranking should be consistent )ith the di$$erence o$ their
evaluation on the ratio scale.
7n the sa#e conte&t, 5ral and 4etanni ,<=>=- propose the opti#i/ation o$
le&icographic criteria )ithout discretisation o$ criteria scales
i
! , )here a
ratio scale is used in order to e&press intensit% o$ pre$erences.
5ther variants o$ the !"A #ethod concerning di$$erent $or#s o$ global
pre$erence are #ainl% $ocused on
9 additional properties o$ the assessed value $unctions, like concavit%
,2espotis and Vopounidis, <==@-I
9 construction o$ $u//% outranking relations based on #ultiple value
$unctions
u
provided b% !"A+s post*opti#alit% anal%sis ,Siskos,
<=>3-.
"he di#ensions o$ the a$ore#entioned !"A #odels a$$ect the
co#putational co#ple&it% o$ the $or#ulated 01s. 7n #ost cases, as noted b%
Fac'uet*0agrG/e and Siskos ,<=>3-, it is pre$erable to solve the dual 01 sue
,. U%/ 1et&o+s 3<
to the structure o$ these 01s. %a)le ,7, su##ari/es the si/e o$ all 01s
presented in the previous sections, )here P and I denote the nu#ber o$
pre$erence and indi$$erence relations respectivel%, considering all possible
pair)ise co#parisons in 8 . Also, it should be noted that 01 ,7*3?- has
, <- 3 m m− binar% variables.
%a)le ,7,. 01 si/e o$ !"A #odels
Model and opti#alit%
criterion
Constraints Jariables
!"A * Min su# o$ errors
,01 7*<D-
<
, <-
n
i
i
m α
·
+ −

<
, <-
n
i
i
m α
·
+ −

!"AS"A6 * Min su# o$
errors ,01 7*3<-
m
<
3 , <-
n
i
i
m α
·
+ −

!"A * Min su# o$ errors
$ro# pair)ise .udg#ents
,01 7*3D-
[ ]
<
< , <- 3 , <-
n
i
i
m m α
·
+ − + −

<
3 , <-
n
i
i
P I α
·
+ + −

!"A 9 Ma& 4endall+s
τ

,01 7*3?-
[ ]
<
< , <- 3 , <-
n
i
i
m m α
·
+ − + −

[ ]
<
, <- 3 , <-
n
i
i
m m α
·
− + −

#.2 Mea!UTA e$hni/'es
5ther techni'ues na#ed #eta*!"A, ai#ed at the i#prove#ent o$ the
value $unction )ith respect to near opti#alit% anal%sis or to its e&ploitation
$or decision support.
2espotis et al. ,<==H- propose to #ini#i/e the dispersion o$ errors
,"cheb%che$$ criterion- )ithin the !"AS"A6+s step D ,see section 3.@-. 7n
case o$ a strictl% positive error
A
, H- < > , the ai# is to investigate the
e&istence o$ near opti#al solutions o$ the 01 ,7*3<- )hich give rankings 8′
such that
A
, , - , , - 8 8 8 8 τ τ ′ > , )ith
A
8
being the ranking corresponding to
the opti#al value $unctions. "he e&perience )ith the #odel ,c$. 2espotis and
Yannacopoulos, <==H- con$ir#s that apart $ro# the total error
A
<
, it is also
the dispersion o$ the individual errors that is crucial $or
A
, , - 8 8 τ . "here$ore,
in the proposed post*opti#alit% anal%sis, the di$$erence bet)een the
#a&i#u#
#a&
, - σ and the #ini#u# error is #ini#i/ed. As $ar as the
individual errors are non*negative, this re'uire#ent can be satis$ied b%
#ini#i/ing the #a&i#u# individual error ,the (

nor#- according to the
$ollo)ing 01
33 '&apter ,
#a&
A
<
#a&
#a&
#a&
B#inC
sub.ect to
all the constraints o$ 01 ,7*3<-
, - , -
, - H
, - H
H
m
" "
"
"
"
a a <
a
a "
σ
σ σ ε
σ σ
σ σ
σ
+ −
·
+

¹
¹
¹
¹
¹
¹
¹
1 + ≤ +
'
¸ ]
¹
¹
− ≥
¹
¹ − ≥ ∀
¹

¹
¹

Lith the incorporation o$ the #odel ,7*3=- in !"AS"A6, the value
$unction assess#ent process beco#es a le&icographic opti#i/ation process.
"hat is, the $inal solution is obtained b% #ini#i/ing successivel% the
<
( and
the (

nor#s.
Another approach concerning #eta*!"A techni'ues re$ers to the
!"AM1 #odels. Eeuthe and Scanella ,<==K, 3HH<- note that the values
given to para#eters
s
and δ in the !"A and !"AS"A6 #ethods,
respectivel%, in$luence the results as )ell as the predictive 'ualit% o$ the
#odels. Sence, in the $ra#e)ork o$ the research b% Srinivasan and Shocker
,<=7@-, the% look $or opti#al values o$
s
and(or δ in the case o$ positive
errors
A
, H- < > , as )ell as )hen !"A gives a su# o$ error e'ual to /ero
A
, H- < · .
7n the post*opti#alit% anal%sis step o$ !"AS"A6 ,see section 3.@-,
!"AM1< #odel #a&i#i/es δ , )hich is the #ini#u# di$$erence bet)een
the global value o$ t)o consecutive re$erence actions. "he na#e o$ the
#odel denotes that, on the basis o$ !"A, #a&i#i/ing δ leads to better
identi$ication $or the relations o$ pre$erence bet)een actions.
Eeuthe and Scanella ,<==K- have also proposed to #a&i#i/e the su#
, - s δ +
in order to stress not onl% the di$$erences o$ utilities bet)een actions,
but also the di$$erences bet)een values at successive bounds. "his #ore
general approach )as na#ed !"AM13. Note that
s
corresponds to the
#ini#u# o$ #arginal value step i;
B
in the !"AS"A6 algorith#. Although
the si#ple addition o$ these para#eters is legiti#ate since both o$ the# are
de$ined in the sa#e value units, Eeuthe and Scanella ,3HH<- note that a
)eighted su# $or#ula #a% also be considered.
#.# So$hasi$ UTA "ehod
Lithin the $ra#e)ork o$ #ulticriteria decision*aid under uncertaint%,
Siskos ,<=>@- developed a speci$ic version o$ !"A ,Stochastic !"A-, in
,. U%/ 1et&o+s 3@
)hich the aggregation #odel to in$er $ro# a re$erence ranking is an additive
utilit% $unction o$ the $or#
< <
, - , - , -
i n
a a ; ;
i i i i
i ;
u $ u $
α
δ
· ·
·
∑∑
0
sub.ect to nor#ali/ation constraints ,7*7-, )here
a
i
δ is the distributional
evaluation o$ action
a
on the i*th criterion, , -
a ;
i i
$ δ is the probabilit% that
the per$or#ance o$ action
a
on the i*th criterion is
;
i
$ , , -
;
i i
u $ is the
#arginal value o$ the per$or#ance
;
i
$ ,
a
0 is the vector o$ distributional
evaluations o$ action
a
and , -
a
u 0 is the global utilit% o$ action
a
,see also
6i$ure ,78-.
- ,
;
i i
$ u
a
i
δ
)
i
δ
;
i
$
- ,
;
i i
$ δ
evaluation
scale !
i
6i$ure ,78. 2istributional evaluation and #arginal value $unction
"his global utilit% is o$ the von Neu#ann*Morgenstern $or# ,c$. 4eene%,
<=>H-, in the case o$ discrete
i
$ , )here
<
, - <
i
a ;
i i
;
$
α
δ
·
·

5$ course, the additive utilit% $unction ,7*@H- has the sa#e properties as
the value $unction
, - , - ,pre$erence-
, - , - ,indi$$erence-
a )
a )
u u a )
u u a )
¹ > ⇔
¹
'
· ⇔
¹
¹
0 0
0 0
f
:
Si#ilarl% to the cases o$ !"A and !"AS"A6 described in sections 3.3*
3.@, the stochastic !"A #ethod disaggregates a ranking o$ re$erence actions
,Siskos and Assi#akopoulos, <=>=-. "he algorith#ic procedure could be
e&pressed in the $ollo)ing )a%
Step 1C
3D '&apter ,
E&press the global e&pected utilities o$ re$erence actions , -
"
a
u 0 ,
<, 3, , " m · K , in ter#s o$ variables
<
, - , - H
; ;
i; i i i i
B u $ u $
+
· − ≥
Step 0C
7ntroduce t)o error $unctions σ
+
and σ

b% )riting the $ollo)ing
e&pressions $or each pair o$ consecutive actions in the ranking
<
<
< <
, , - , - , - , -
, - , - , -
"
"
a
" " " "
a
" "
a a u a a
u a a
σ σ
σ σ
+
+ −
+
+ −
+ +
∆ · − +
− + −
0
0
Step 3C
Solve the linear progra# ,7*3<- b% using $or#ulae ,7*@@- and ,7*@D-.
Step 4C
"est the e&istence o$ #ultiple or near opti#al solutions.
5$ course, the ideas e#plo%ed in all variants o$ the !"A #ethod are also
applicable in the sa#e )a% in the case o$ the stochastic !"A.
#.+ UTA!ype soring "ehods
"he e&tension o$ the !"A #ethod in the case o$ a discri#inant anal%sis
#odel )as $irstl% discussed b% Fac'uet*0agrG/e and Siskos ,<=>3-. "he ai#
is to in$er
u
$ro# assign#ent e&a#ples in the conte&t o$ proble#atic M ,c$.
6o%, <=>?-. 7n the presence o$ t)o classes, i$ the #odel is )ithout errors, the
$ollo)ing ine'ualities #ust hold
[ ]
[ ]
<
3
, -
, -
o
o
a / u a u
a / u a u
¹ ∈ ⇔ ≥
¹
'
∈ ⇔ <
¹
¹
g
g
)ith
o
u being the level o$ acceptance(re.ection )hich #ust be $ound in order
to distinguish the set o$ accepted actions called
<
/ and the set o$ re.ected
actions called
3
/ .
7ntroducing the error variables
, - a σ
,
8
a / ∈ , the ob.ective is to
#ini#i/e the su# o$ deviations $ro# the threshold
o
u $or the ill classi$ied
actions ,see 6i$ure ,7F-. Sence,
, - u g
can be esti#ated b% #eans o$ the 01
,. U%/ 1et&o+s 3?
[ ]
[ ]
H <
<
H 3
<
<
A
<
B#inC , -
sub.ect to
, - , - H
, - , - H
, - , - and
, - <
8
a /
n
i i
i
n
i i
i
; ;
i i i i i
i i
i
6 a
u $ a u a a /
u $ a u a a /
u $ u $ s i ;
u $
σ
σ
σ

·
·
+
·
·
− + ≥ ∀ ∈
− − ≤ ∀ ∈
− ≥ ∀
·



A H
, - H, H, , - H, , - H , and
n
;
i i i i 8
u $ u u $ a a / i ; σ
¹
¹
¹
¹
¹
¹
¹
¹
¹
'
¹
¹
¹
¹
¹
¹
¹ · ≥ ≥ ≥ ∀ ∈ ∀
¹
¹

H
u H <
, - u g
, - a σ
/
<
/
3
6i$ure ,7F. 2istribution o$ the actions /< and /3 on u,g- ,Fac'uet*0agrG/e and Siskos, <=>3-
7n the general case, the 2M+s evaluation is e&pressed in ter#s o$ a
classi$ication o$ the re$erence alternatives into ho#ogenous ordinal groups
< 3 E
/ / / f f K f
,i.e. group
<
/ includes the #ost pre$erred alternatives,
)hereas group E
/
includes the least pre$erred ones-. Lithin this conte&t, the
assessed additive value #odel )ill be consistent )ith the 2M+s global
.udg#ent, i$ the $ollo)ing conditions are satis$ied
[ ]
[ ]
[ ]
< <
<
<
, -
, - , 3, @, , <-
, -
l l l
E E
u a u a /
u u a u a / l E
u a u a /


¹ ≥ ∀ ∈
¹
¹
≤ < ∀ ∈ · −
'
¹
< ∀ ∈
¹
¹
g
g
g
K
3K '&apter ,
)here < 3 < E
u u u

> > > K
are thresholds de$ined in the global value scale
BH,<C
to discri#inate the groups, and
l
u is the lo)er bound o$ group
l
/ .
"his approach is na#ed !"A27S #ethod ,!"ilitNs Additives
27Scri#inantes- and is presented b% 2evaud et al. ,<=>H-, Fac'uet*0agrG/e
,<==?-, Vopounidis and 2ou#pos ,<==7-, Vopounidis and 2ou#pos ,3HH<-,
2ou#pos and Vopounidis ,3HH3-. Si#ilarl%, to the !"AS"A6 #ethod, t)o
error variables are e#plo%ed in the !"A27S #ethod to #easure the
di$$erences bet)een the #odel+s results and the prede$ined classi$ication o$
the re$erence alternatives. "he additive value #odel is developed to
#ini#i/e these errors using a linear progra##ing $or#ulation o$ t%pe ,7*
@K-. 7n this case, the t)o t%pes o$ errors are de$ined as $ollo)s
<. { } #a& H, B , -C
" l "
u u a σ
+
· − g
" l
a / ∀ ∈
, <, 3, <- l E · − K
represents the
error associated )ith the violation o$ the lo)er bound
l
u o$ a group
l
/ b% an alternative
" l
a / ∈ ,
3. { }
<
#a& H, B , -C
" " l
u a u σ


· − g
" l
a / ∀ ∈
, 3, @, - l E · K
represents the
error associated )ith the violation o$ the upper bound
< l
u

o$ a group
l
/ b% an alternative
" l
a / ∈ .
6ecentl%, several ne) variants o$ the original !"A27S #ethod have
been proposed ,!"A27S 7, 77, 777- to consider di$$erent opti#alit% criteria
during the develop#ent o$ the additive value classi$ication #odel
,Vopounidis and 2ou#pos, <==7I Vopounidis and 2ou#pos, 3HH<I
2ou#pos and Vopounidis, 3HH3-. "he !"A27S 7 #ethod considers both the
#ini#i/ation o$ the classi$ication errors, as )ell as the #a&i#i/ation o$ the
distances o$ the correctl% classi$ied alternatives $ro# the value thresholds.
"he ob.ective in the !"A27S 77 #ethod is to #ini#i/e the nu#ber o$
#isclassi$ied alternatives, )hereas !"A27S 777 co#bines the #ini#i/ation
o$ the #isclassi$ied alternatives )ith the #a&i#i/ation o$ the distances o$ the
correctl% classi$ied alternatives $ro# the value thresholds.
7n the sa#e conte&t, Vopounidis and 2ou#pos ,3HHHa- proposed the
MS27S #ethod ,Multi*group Sierarchical 27Scri#ination- e&tending the
pre$erence disaggregation anal%sis $ra#e)ork o$ the !"A27S #ethod in
co#ple& sorting(classi$ication proble#s involving #ultiple*groups. MS27S
addresses sorting proble#s through a hierarchical ,se'uential- procedure
starting b% discri#inating group
<
/ $ro# all the other groups 3 @
O , , , P
E
/ / / K
, and then proceeding to the discri#ination bet)een the alternatives
belonging to the other groups. At each stage o$ this se'uential(hierarchical
process, t)o additive value $unctions are developed $or the classi$ication o$
the alternatives. Assu#ing that the classi$ication o$ the alternatives should be
#ade into E ordered classes < 3 E
/ / / f f K f
,
3, <- E −
additive value
$unctions are developed. "hese value $unctions have the $ollo)ing additive
$or#
,. U%/ 1et&o+s 37
<
<
, - , -
, - , -
n
l li i
i
n
l li i
i
u u $
u u $
·
·
¹
·
¹
¹
'
¹
·
¹
¹


g
g
: :
)here
l
u #easures the value $or the 2M o$ a decision to assign an
alternative into group
l
/ , )hereas the
l
u
:
corresponds to the classi$ication
into the set o$ groups < 3
O , , , P
l l l E
/ / / /
+ +
·
:
K
and both $unctions are
nor#ali/ed in the interval
BH,<C
.
"he rules used to per$or# the classi$ication o$ the alternatives have the
$ollo)ing $or#
< < <
3 3 3
< , <- <
i$ , - , - then
else i$ , - , - then

else i$ , - , - then
else
" " "
" " "
E " E " " E
" E
u a u a a /
u a u a a /
u a u a a /
a /
− − −
¹
> ∈
¹
> ∈
¹
¹
'
¹
> ∈
¹
¹

¹
:
:
:
KKKKKKKKKKKKK
"he develop#ent o$ all value $unctions in the MS27S #ethod is
per$or#ed through the solution o$ three #athe#atical progra##ing
proble#s at each stage l o$ the discri#ination process
<, 3, < l E · − K
.
7nitiall%, a 01 is solved to #ini#i/e the #agnitude o$ the classi$ication
errors ,in distance ter#s si#ilarl% to the !"A27S approach-. "hen, a #i&ed*
integer 01 is solved to #ini#i/e the total nu#ber o$ #isclassi$ications
a#ong the #isclassi$ications that occur a$ter the solution o$ the initial 01,
)hile retaining the correct classi$ications. Finall%, a second 01 is solved to
#a&i#i/e the clarit% o$ the classi$ication obtained $ro# the solutions o$ the
previous 01s.
#.1 Oher )arians and e,ensions
7n all previous approaches, the value $unction )as built in a one*step
process b% $or#ulating a 01 that re'uires onl% the 2M+s global pre$erences.
7n so#e cases, ho)ever, it )ould be #ore appropriate to build such a
$unction $ro# a t)o*step 'uestioning process, b% dissociating the
construction o$ the #arginal value $unctions and the assess#ent o$ their
respective scaling constants.
7n the $irst step, the various #arginal value $unctions are built outside the
!"A algorith#. "hese $unctions #a% be $acilitated, $or instance, b%
proposing speci$ic para#etrical #arginal value $unctions to the 2M and
asking hi#(her to choose the one that #atches his(her pre$erences on that
3> '&apter ,
speci$ic criterion. "hose $unctions should be nor#ali/ed according to ,7*D-
conditions. Generall%, the approaches applied in this construction step are
a- techni'ues based on MA!" theor% and described b% 4eene% and
6ai$$a ,<=7K-, and 4lein et al. ,<=>?-,
b- the MACEE"S #ethod ,Eana e Costa and Jansnick, <==D, <==7I
Eana e Costa et al., 3HH<-,
c- the Wuasi*!"A #ethod b% Eeuthe et al. ,3HHH-, that uses :recursive
e&ponential; #arginal value $unctions, and
d- the M772AS s%ste# ,see section D- that co#bines arti$icial
intelligence and visual procedures in order to e&tract the 2M+s
pre$erences ,Siskos et al., <===-.
7n the second step, a$ter the assess#ent o$ these value $unctions, the 2M
is asked to give a global ranking o$ alternatives in a si#ilar )a% as in the
basic !"A #ethod. Fro# this in$or#ation, the proble# #a% be $or#ulated
via a 01, in order to assess onl% the )eighting $actors
i
p o$ the criteria
,scaling constants o$ criteria-. "hrough this approach, initiall% na#ed !"A
77 #odel ,Siskos, <=>H-, $or#ula ,7*<<- beco#es
{ }
<
, , - B , -C B , -C
, - , - , - , -
n
i i i i i
i
a ) p u $ a u $ )
a a ) ) σ σ σ σ
·
+ − + −
∆ · −
− + + −

and the 01 ,7*<D- is #odi$ied as $ollo)s
<
B#inC , - , -
sub.ect to
, , - i$
, , - H i$
<
H, , - H, , - H ,
8
a /
n
i
i
i 8
6 a a
a ) a )
a ) a )
p
p a a a / i
σ σ
δ
σ σ
+ −

·
+ −
¹ 1 · +
¸ ]
¹
¹
¹
¹
∆ ≥
¹
'
∆ ·
¹
¹
·
¹
¹
¹
≥ ≥ ≥ ∀ ∈ ∀
¹


f
:
"he #ain principles o$ the !"A #ethods are also applicable in the
speci$ic $ield o$ #ultiob.ective opti#i/ation, #ainl% in the $ield o$ linear
progra##ing )ith #ultiple ob.ective $unctions. For instance, in the classical
#ethods o$ Geo$$rion et al. ,<=73- and Vionts and Lallenius ,<=7K-, the
)eights o$ the linear co#binations o$ the ob.ectives are in$erred locall% $ro#
trade9o$$s or pair)ise .udg#ents given b% the 2M at each iteration o$ the
#ethods. "hus, these #ethods e&ploit in a direct )a% the 2M+s value
$unctions and seek the best co#pro#ise solution through successive
#a&i#i/ation o$ these assessed value $unctions.
Ste)art ,<=>7- proposed a procedure o$ pruning the decision alternatives
using the !"A #ethod. 7n this approach a se'uence o$ alternatives is
,. U%/ 1et&o+s 3=
presented to the 2M, )ho places each ne) presented alternative in rank
order relative to the earlier alternatives evaluated. "his ranking o$ ele#ents
in a subset o$ the decision space is used to eli#inate other alternatives $ro#
$urther consideration. 7n the sa#e conte&t, Fac'uet*0agrG/e et al. ,<=>7-
developed a disaggregation #ethod, si#ilar to !"A, to assess a )hole value
$unction o$ #ultiple ob.ectives $or linear progra##ing s%ste#s. "his
#ethodolog% enables to $ind co#pro#ise solutions and is #ainl% based on
the $ollo)ing steps
<. Generation o$ a li#ited subset o$ $easible e$$icient solutions as
representative as possible o$ the e$$icient set.
3. Assess#ent o$ an additive value $unction using 16EFCA0C s%ste#
,see section D-.
@. 5pti#i/ation o$ the additive value $unction on the original set o$
$easible alternatives.
Also, Siskos and 2espotis ,<=>=-, in the conte&t o$ !"A*based
approaches in #ultiob.ective opti#i/ation proble#s, proposed the
A2E0A7S #ethod. "his approach re$ers to an interactive #ethod that uses
!"A iterativel%, in order to opti#i/e an additive value $unction )ithin the
$easible region de$ined on the basis o$ the satis$action levels and deter#ined
in each iteration.
#.2 Oher disaggregaion "ehods
"he #ain principles o$ the aggregation*disaggregation approach #a% be
co#bined )ith outranking relation #ethods. "he #ost i#portant e$$orts
concern the proble# o$ deter#ining the values o$ several para#eters )hen
using these #ethods. "he set o$ these para#eters is used to construct a
pre$erence #odel )ith )hich the 2M accepts as a )orking h%pothesis in the
decision aid stud%. Assu#ing that the 2M is able to give e&plicitl% the
values o$ each para#eter is not realistic in several real*)orld applications.
7n this $ra#e)ork, the E0ECCA0C s%ste# has been developed ,4iss et
al., <==D-, )hich esti#ates indirectl% the para#eters o$ the E0EC"6E 77
#ethod. "he process is based on the 2M+s responses to 'uestions o$ the
s%ste# regarding his(her global pre$erences.
Further#ore, concerning proble#atic 3, several approaches consist in
in$erring the para#eters o$ E0EC"6E "67 through holistic in$or#ation on
2M+s .udg#ents. "hese approaches ai# at substituting assign#ent e&a#ples
$or direct elicitation o$ the #odel para#eters. !suall%, the values o$ these
para#eters are in$erred through a regression*t%pe anal%sis on assign#ent
e&a#ples.
Mousseau and Slo)inski ,<==>- propose an interactive aggregation*
disaggregation approach that in$ers E0EC"6E "67 para#eters
@H '&apter ,
si#ultaneousl% starting $ro# assign#ent e&a#ples. 7n this approach, the
deter#ination o$ the para#eters+ values ,e&cept the veto thresholds- that best
restore the assign#ent e&a#ples is $or#ulated through a nonlinear
opti#i/ation progra#.
Several e$$orts have tried to overco#e the li#itations o$ the
a$ore#entioned approach ,co#putational di$$icult%, esti#ation o$ the veto
threshold-
a- Mousseau et al. ,3HHHaI 3HHHb- consider the subproble# o$ the
deter#ination o$ the )eights onl%, assu#ing that the thresholds and
categor% li#its have been $i&ed. "his leads to $or#ulate a linear
progra# ,rather than nonlinear in the global in$erence #odel-.
"hrough e&peri#ental anal%sis, the% sho) that this approach is able to
in$er )eights that restore in a stable )a% the assign#ent e&a#ples and
it is also able to identi$% possible inconsistencies in these assign#ent
e&a#ples.
b- 2ou#pos and Vopounidis ,3HH3- use linear progra##ing
$or#ulations in order to esti#ate all the para#eters o$ the outranking
relation classi$ication #odel. So)ever, in this approach, the
para#eters are esti#ated se'uentiall% rather than through a global
in$erence process. "hus, the proposed #ethodolog% does not speci$%
the opti#al para#eters o$ the outranking relation ,i.e. the ones that
lead to a global #ini#u# o$ the classi$ication error-. "here$ore, the
results o$ this approach ,:reasonable; speci$ication o$ the para#eters-
serve rather as a basis $or a thorough decision*aid process.
"he proble# o$ robustness and sensitivit% anal%sis, through the e&tension
o$ the previous research e$$orts is discussed b% 2ias et al. ,3HH3-. "he%
consider the case )here the 2M can not provide e&act values $or the
para#eters o$ the E0EC"6E "67 #ethod, due to uncertain, i#precise or
inaccuratel% deter#ined in$or#ation, as )ell as $ro# lack o$ consensus
a#ong the#. "he proposed #ethodolog% co#bines the $ollo)ing
approaches
<. "he $irst approach in$ers the value o$ para#eters $ro# assign#ent
e&a#ples provided b% the 2M, as an elicitation aid.
3. "he second approach considers a set o$ constraints on the para#eter
values re$lecting the i#precise in$or#ation that the 2M is able to
provide.
7n the conte&t o$ ordinal regression anal%sis, the M!SA #ethod has been
developed in order to #easure and anal%/e custo#er satis$action ,Siskos et
al., <==>I Grigoroudis and Siskos, 3HH3-. "he #ethod is used $or the
assess#ent o$ a set o$ #arginal satis$action $unctions in such a )a% that the
global satis$action criterion beco#es as consistent as possible )ith
,. U%/ 1et&o+s @<
custo#er+s .udg#ents. "hus, the #ain ob.ective o$ the #ethod is the
aggregation o$ individual .udg#ents into a collective value $unction.
"he M!SA #ethod assesses global and partial satis$action $unctions
A
G

and
A
i
H respectivel%, given custo#ers+ ordinal .udg#ents G and
i
H ,$or the
i*th criterion-. "he ordinal regression anal%sis e'uation has the $ollo)ing
$or#
A A
<
n
i i
i
G ) H σ σ
+ −
·
· − +

%
)here
A
G
%
is the esti#ation o$ the global value $unction
A
G
,
n
is the nu#ber
o$ criteria,
i
) is a positive )eight o$ the i*th criterion, σ
+
and σ

are the
overesti#ation and the underesti#ation errors, respectivel%, and the value
$unctions
A
G
and
A
i
H are nor#ali/ed in the interval BH,<HHC.
Si#ilarl% to the !"AS"A6 algorith#, the $ollo)ing trans$or#ation
e'uations are used
A < A
A < A
$or < 3 <
$or < 3 < and < 3
m m
m
" "
i" i i i i i
< y y m= * *...*2
B ) @ ) @ " = * *...*2 i = * *...*n
+
+
¹ · − −
¹
'
· − −
¹
¹
)here
Am
y is the value o$ the
m
y satis$action level,
A"
i
@ is the value o$ the
"
i
@ satis$action level, and
α
and
i
α are the nu#ber global and partial
satis$action levels.
According to the previous de$initions and assu#ptions, the M!SA
esti#ation #odel can be )ritten in a 01 $or#ulation, as $ollo)s
[ ]
*
<
< <
< < <
<
<
<
< <
#in
sub.ect to
H $or < 3
<HH
<HH
, , ,
; ;
i
i
1
; ;
;
@ y n
i" m ; ;
i " m
2
m
m
2 n
i"
i "
m i" ; ;
6
B < ; = * * *1
<
B
< B * m*i* ; "
σ σ
σ σ
σ σ
+
·
− −
+ −
· · ·

·

· ·
+ −
¹
· +
¹
¹
¹
¹
¹
− − + ·
¹
¹
'
¹
·
¹
¹
¹
·
¹
¹
∀ ¹
¹

∑∑ ∑

∑∑
K
)here 1 is the si/e o$ the custo#er sa#ple, and
;
i
@ and
;
y are the ;*th
level on )hich variables
i
H and G are esti#ated ,i.e. global and partial
satis$action .udg#ents o$ the ;*th custo#er-. "he M!SA #ethod includes
also a post*opti#alit% anal%sis stage, si#ilarl% to step D o$ the !"AS"A6
algorith#.
An anal%tical develop#ent o$ the #ethod and the provided results is
given b% Grigoroudis and Siskos ,3HH3-, )hile the presentation o$ the
@3 '&apter ,
M!SA 2SS can be $ound in Grigoroudis et al. ,3HHH- and Grigoroudis and
Siskos ,3HH@-.
"he proble# o$ building non*additive utilit% $unctions #a% also be
considered in the conte&t o$ aggregation*disaggregation approach. A
characteristic case re$ers to positive interaction ,s%nerg%- or negative
interaction a#ong criteria ,redundanc%-. ")o or #ore criteria are s%nergic
,redundant- )hen their .oint )eight is #ore ,less- than the su# o$ the
)eights given to the criteria considered singularl%.
7n order to represent interaction a#ong criteria, so#e speci$ic
$or#ulations o$ the utilit% $unctions e&pressed in ter#s o$ $u//% integrals
have been proposed ,Muro$ushi and Sugeno, <=>=I Grabisch, <==KI
Marichal and 6oubens, 3HHH-. 7n this conte&t, Angilella et al. ,3HH@- propose
a #ethodolog% that allo)s including additional in$or#ation such as an
interaction a#ong criteria. "he #ethod ai#s at searching a utilit% $unction
representing the 2M+s pre$erences, )hile the resulting $unctional $or# is a
speci$ic $u//% integral ,Cho'uet integral-. As a result, the obtained )eights
#a% be interpreted as the :i#portance; o$ coalitions o$ criteria, e&ploiting
the potential interaction bet)een criteria. "he #ethod can also provide the
#arginal utilit% $unctions relative to each one o$ the considered criteria,
evaluated on a co##on scale, as a conse'uence o$ the i#ple#ented
#ethodolog%.
"he general sche#e o$ the disaggregation philosoph% is also e#plo%ed in
other approaches, including rough sets ,1a)lak, <=>3I Slo)inski, <==?I
2i#itras, et al., <===I Varas, 3HHH-, #achine learning ,Wuinlan, <=>K- and
neural net)orks ,Malakooti and Vhou, <==DI Sta# et al., <==K-. All these
approaches are used to in$er so#e $or# o$ decision #odel ,a set o$ decision
rules or a net)ork- $ro# given decision results involving assign#ent
e&a#ples, ordinal or #easurable .udg#ents.
+. A((3ICATIONS AND UTA!4ASED DSS
"he #ethods presented in the previous sections adopt the aggregation*
disaggregation approach. "his approach constitutes a basis $or the interaction
bet)een the anal%st and the 2M, )hich includes
9 the consistenc% bet)een the assessed pre$erence #odel and the a priori
pre$erences o$ the 2M,
9 the assessed values ,values, )eights, utilities, X-, and
9 the overall evaluation o$ potential actions ,e&trapolation output-.
A general interaction sche#e $or this decision support process is given in
6i$ure ,710.
,. U%/ 1et&o+s @@
Start
1roble# $or#ulation
Criteria #odelling
E&pression o$ 2MYs global
pre$erences on a re$erence set
Assess the 2MYs pre$erence #odel
7s there a satis$actor% consistenc%
bet)een the 2MYs pre$erences and
the ones resulted b% the #odel8
7s there an% intention o$ #odi$%ing
the pre$erence #odel8
7s there an% intention o$ #odi$%ing
the 2MYs pre$erences8
7s there an% intention o$ #odi$%ing
the criteria #odelling or the
proble# $or#ulation8
"he pre$erence #odel is .udged
unacceptable
E&trapolation on the )hole
set o$ alternatives
7s the pre$erence #odel
accepted8
End
N5
YES
N5
N5
N5
End
YES
YES
YES
YES
N5
6i$ure ,710. Si#pli$ied decision support process based on disaggregation approach ,Fac'uet*
0agrG/e and Siskos, 3HH<-
Several decision support s%ste#s ,2SSs-, based on the !"A #odel and
its variants, have been developed on the basis o$ disaggregation #ethods.
"hese s%ste#s include
@D '&apter ,
a- "he 16EFCA0C s%ste# ,Fac'uet*0agrG/e, <==H- is a 2SS $or
interactive assess#ent o$ pre$erences using holistic .udg#ents. "he
interactive process includes the classical aggregation phase )here the
2M is asked to esti#ate directl% the para#eters o$ the #odel ,i.e.
)eights, trade*o$$s, etc.-, as )ell as the disaggregation phase )here the
2M is asked to e&press his(her holistic .udg#ents ,i.e. global
pre$erence order on a subset o$ the alternatives- enabling an indirect
esti#ation o$ the para#eters o$ the #odel.
b- M7N56A ,Multicriteria 7nteractive 5rdinal 6egression Anal%sis- is a
#ulticriteria interactive 2SS )ith a )ide spectru# o$ supported
decision #aking situations ,Siskos et al., <==@, <==D-. "he core o$ the
s%ste# is based on the !"AS"A6 #ethod and it uses special
interaction techni'ues in order to guide the 2M to reach a consistent
pre$erence s%ste#.
c- M772AS ,Multicriteria 7nteractive 7ntelligence 2ecision Aiding
S%ste#- is an interactive 2SS that i#ple#ents the e&tended !"A 77
#ethod ,Siskos et al., <===-. 7n the $irst step o$ the decision*aiding
process, the s%ste# assess the 2M+s value $unctions, )hile in the ne&t
step, the s%ste# esti#ates the 2M+s pre$erence #odel $ro# his(her
global pre$erences on a re$erence set o$ alternative actions. "he s%ste#
uses Arti$icial 7ntelligence and Jisual techni'ues in order to i#prove
the user inter$ace and the interactive process )ith the 2M ,6i$ure ,7
11-.
d- "he !"A 10!S so$t)are ,4ostko)ski and Slo)inski, <==KI
http(())).la#sade.dauphine.$r(english(so$t)are.ht#lZutaQ- is an
i#ple#entation o$ the !"A #ethod, )hich allo)s the user to #odi$%
interactivel% the #arginal value $unctions )ithin li#its $ollo)ing $ro#
a sensitivit% anal%sis o$ the $or#ulated ordinal regression proble#.
2uring all these #odi$ications, a $riendl% graphical inter$ace helps the
2M to reach an accepted pre$erence #odel.
e- M!S"A62 ,Multicriteria !tilit%*based Stochastic Aid $or 6anking
2ecisions- is an interactive 2SS developed b% Eeuthe and Scannella
,<===-, )hich incorporates several variants o$ the !"A #ethod. "he
s%ste# provides several visual tools in order to structure the 2M+s
pre$erences to a speci$ic proble# ,see also Siskos, 3HH3-. "he
interactive process )ith the 2M contains the $ollo)ing #ain steps
proble# structuring, pre$erence 'uestionnaire, opti#i/ation solver*
para#eter co#puting, $inal results ,$ull rankings and graphs-.
,. U%/ 1et&o+s @?
2A"A S!ESYS"EM
 2ata o$ the proble# ,actions*criteria,
characteristics*per$or#ance table-
 2M+s pre$erences ,re$erence set, value
$unction, ranking-
 Assessed pre$erence #odel ,)eights,
over(underesti#ation errors-
 2ata o$ trade*o$$ anal%sis ,ranking, ne)
value $unctions, )eights, over(
underesti#ation errors-
M52E0 S!ESYS"EM
 !"A 77
 0inear progra##ing techni'ues
 2o#ination relations
 Cluster anal%sis
 Jalue $unction esti#ation #ethods
 "rade*odd anal%sis #ethod
 Co#putations ,#arginal utilities, global
utilities, assessed ranking,
e&trapolation-
27A05G!E S!ESYS"EM
 Lindo)s
 "ables
 Edit $ields
 Messages
 Graphs
 Jisual tools
 1ictures
2EC7S75N ANA0YS"
2EC7S75N MA4E6
Jisual s%ste#
A7 techni'ues o$
#odel subs%ste#
A7 techni'ues o$
dialogue subs%ste#
6i$ure ,711. Co#ponents o$ M772AS s%ste# ,Siskos, et al., <===-
!"A #ethods have also been used in several )orks $or con$lict
resolution in #ulti*actor decision situations ,Fac'uet*0agrG/e and Shakun,
<=>DI Eui, <=>7I Matsatsinis and Sa#aras, 3HH<-. 7n the sa#e conte&t, the
ME27A"56 s%ste# )as developed ,Farke et al., <=>>I Shakun, <=>>I
Shakun, <==<-, )hich is a negotiation support s%ste# based on Evolutionar%
S%ste#s 2esign ,ES2- and database*centered i#ple#entation. ES2
visuali/es negotiations as a collective process o$ searching $or designing a
#utuall% acceptable solution. 1articipants are seen as pla%ing a d%na#ical
di$$erence ga#e in )hich a coalition o$ pla%ers is $or#ed, i$ it can achieve a
set o$ agreed upon goals. 7n ME27A"56, negotiations are supported b%
consensus seeking through e&change o$ in$or#ation and, )here consensus is
inco#plete, b% co#pro#ise. 7t assists in consensus seeking b% aiding the
pla%ers to build a group .oint proble# representation o$ the negotiations*in
e$$ect, .oint #appings $ro# control space to goal space ,and through
#arginal utilit% $unctions- to utilit% space. 7ndividual #arginal utilit%
$unctions are esti#ated b% appl%ing the !"A #ethod. 1la%ers can arrive to a
co##on coalition utilit% $unction through e&change o$ in$or#ation and
negotiation until pla%ers+ #arginal utilit% $unctions are identical. 7n addition
@K '&apter ,
to e&changing in$or#ation and negotiating to e&pand targets, pla%ers can
consider the use o$ a&io#s to contract the $easible region.
7n the area o$ intelligent #ulticriteria 2SSs, the MA64ER s%ste# has
been proposed b% Siskos and Matsatsinis ,<==@-, Matsatsinis and Siskos
,<===-, Matsatsinis and Siskos ,3HH@-. "he s%ste# includes the !"AS"A6
algorith# and is used $or the ne) product develop#ent process. 7t acts as a
consultant $or #arketers, providing visual support to enhance understanding
and to overco#e lack o$ e&pertise. "he data bases o$ the s%ste# are the
results o$ consu#er surve%s, as )ell as $inancial in$or#ation o$ the
enterprises involved in the decision #aking process. "he s%ste#+s #odel
base enco#passes statistical anal%sis, pre$erence anal%sis, and consu#er
choice #odels. 6i$ure ,710 presents a general #ethodological $lo)chart o$
the s%ste#. Also, MA64ER incorporates partial kno)ledge bases to support
decision #akers in di$$erent stages o$ the product develop#ent process. "he
s%ste# incorporates three partial e&pert s%ste#s, $unctioning independentl%
o$ each other "he% use the $ollo)ing kno)ledge bases $or the
9 selection o$ data anal%sis #ethod,
9 selection o$ brand choice #odel, and
9 evaluation o$ the $inancial status o$ enterprises.
Further#ore, an intelligent )eb*based 2SS, na#ed 27M7"6A, has been
developed b% Matsatsinis and Siskos ,3HH<-. "he s%ste# is a consu#er
surve%*based 2SS, $ocusing on the decision*aid process $or agricultural
product develop#ent. Eesides the i#ple#entation o$ the !"AS"A6 #ethod
in the pre$erence anal%sis #odule, the 27M7"6A s%ste# co#prises several
statistical anal%sis tools and consu#er choice #odels. "he s%ste# provides
visual support to the 2M ,agricultural cooperatives, agribusiness $ir#s, etc.-
$or several co#ple& tasks, such as
9 evaluation o$ current and potential #arket shares,
9 deter#ination o$ the appropriate co##unication and penetration
strategies, based on consu#er attitudes and belie$s,
9 ad.ust#ent o$ the production according to product+s de#and, and
9 detection o$ the #ost pro#ising #arkets.
7n the sa#e conte&t, ne) research e$$orts have co#bined !"A*based
2SSs )ith intelligent agents+ technolog%. 7n general, the proposed
#ethodologies engage the !"A #odels in a #ulti*agent architecture in order
to assess the 2M+s pre$erence s%ste#. "hese research e$$orts include #ainl%
the $ollo)ing
a- An intelligent agent*based 2SS, $ocusing on the deter#ination o$
product penetration strategies has been developed b% Matsatsinis et al.
,<===, 3HHH, 3HH<-. "he s%ste# i#ple#ents an original consu#er*
based #ethodolog%, in )hich intelligent agents operate in a $unctional
and a structural level, si#ultaneousl%. "ask, in$or#ation and inter$ace
,. U%/ 1et&o+s @7
agents are included in the $unctional level in order to coordinate,
collect necessar% in$or#ation and co##unicate )ith the 2M.
0ike)ise, the structural level includes ele#entar% agents based on a
generic reusable architecture and co#ple& agents )hich ai# to the
develop#ent o$ a d%na#ical agent organi/ation in a recursive )a%.
Forecasting #odels
Market characterictics
Methods o$ data
anal%sis
Consu#ers+
characteristics
!"AS"A6
Multicriteria anal%sis
Market trends
Consu#er behavior
Erand choice #odels
Si#ulations
Esti#ation o$ #arket shares
Ne) product design
Surve%
Creation o$ databases
Market characterictics
Selection o$ penetration
strateg%
6i$ure ,710. Methodological $lo)chart o$ MA64ER ,Matsatsinis and Siskos, <===-
@> '&apter ,
b- A #ulti*agent architecture is proposed b% Manouselis and Matsatsinis
,3HH<- $or #odeling electronic consu#er+s behavior. "he
i#ple#entation o$ the s%ste# re$ers to electronic #arketplaces and
incorporates a step*b%*step #ethodolog% $or intelligent s%ste#s
anal%sis and design, used in the particular decision*aid process. "he
s%ste# develops consu#er behavioral #odels $or the purchasing and
negotiation process adopting additional operations research tools and
techni'ues. "he presented application re$ers to the case o$ 7nternet
radio.
c- "he AgentAllocator s%ste# ,Matsatsinis and 2elias, 3HH@- i#ple#ents
the !"A 77 #ethod in the task allocation proble#. "hese proble#s are
ver% co##on to an% #ulti*agent s%ste# in the conte&t o$ Arti$icial
7ntelligence. "he s%ste# is an intelligent agent 2SS, )hich allo)s the
2M to #odel his(her pre$erences in order to reach and e#plo% the
opti#al allocation plan.
"he need to co#bine data and kno)ledge in order to solve co#ple& and
ill*structured decision proble#s is a #a.or concern in the #odern #arketing*
#anage#ent science. Matsatsinis ,3HH3- has proposed a 2SS that
i#ple#ents the !"AS"A6 algorith# along )ith rule*induction data #ining
techni'ues. "he #ain ai# o$ the s%ste# is to derive and appl% a set o$ rules
that relate the global and the #arginal value $unctions. A co#parison
bet)een the original and the rule*based global values is used in the validit%
and stabilit% anal%sis o$ the proposed #ethodolog%.
Further#ore, in the area o$ $inancial #anage#ent, a variet% o$ !"A*
based 2SSs has been developed, including #ainl% the $ollo)ing s%ste#s
a- "he F7NEJA s%ste# ,Vopounidis et al., <==K- is a #ulticriteria
kno)ledge*based 2SS developed $or the assess#ent o$ corporate
per$or#ance and viabilit%. "he s%ste# i#ple#ents #ultivariate
statistical techni'ues ,e.g. principal co#ponents anal%sis-, e&pert
s%ste#s technolog%, and the !"AS"A6 #ethod to provide integrated
support in evaluating the corporate per$or#ance.
b- "he F7NC0AS s%ste# ,Vopounidis and 2ou#pos, <==>- is a
#ulticriteria 2SS developed to stud% $inancial decision*#aking
proble#s in )hich a classi$ication ,sorting- o$ the alternatives is
re'uired. "he present $or# o$ the s%ste# is devoted to corporate credit
risk assess#ent, and it can be used to develop classi$ication #odels to
assign a set o$ $ir#s into prede$ined credit risk classes. "he anal%sis
per$or#ed b% the s%ste# is based on the $a#il% o$ the !"A27S
#ethods.
c- "he 7NJES"56 s%ste# ,Vopounidis and 2ou#pos, 3HHHb- is
developed to stud% proble#s related to port$olio selection and
#anage#ent. "he s%ste# i#ple#ents the !"A27S #ethod, as )ell as
,. U%/ 1et&o+s @=
goal progra##ing techni'ues to support port$olio #anagers and
investors in their dail% practice.
d- "he 16EF27S s%ste# ,Vopounidis and 2ou#pos, 3HHHc- is a
#ulticriteria 2SS developed to address classi$ication proble#s. "he
s%ste# i#ple#ents a series o$ pre$erence disaggregation anal%sis
techni'ues, na#el% the $a#il% o$ the !"A27S #ethods, in order to
develop an additive utilit% $unction to be used $or classi$ication
purposes
Finall%, as presented in section @.?, Siskos and 2espotis ,<=>=- have
developed the A2E0A7S s%ste# )hich is designed to decision*aid in
#ultiob.ective linear progra##ing ,M501- proble#s.
5ver the past t)o decades !"A*based #ethods have been applied in
several real*)orld decision*#aking proble#s $ro# the $ields o$ $inancial
#anage#ent, #arketing, environ#ental #anage#ent, as )ell as hu#an
resources #anage#ent, as presented in %a)le ,71. "hese applications have
provided insight on the applicabilit% o$ pre$erence disaggregation anal%sis in
addressing real*)orld decision proble#s and its e$$icienc%.
%a)le ,78. 7ndicative applications o$ the !"A #ethods
Field Scope 6e$erence
Financial
#anage#ent
Jenture capital evaluation Siskos and Vopounidis ,<=>7-
1ort$olio selection and #anage#ent Surson and Vopounidis ,<==7-
Vopounidis et al. ,<===-
Eusiness $ailure prediction Vopounidis ,<=>7-
Vopounidis and 2ou#pos ,<===-
Eusiness $inancing Siskos et al. ,<==D-
Vopounidis et al. ,<==K-
Vopounidis and 2ou#pos ,<==>-
Countr% risk assess#ent Cosset et al. ,<==3-
5ral et al. ,<==3-
Vopounidis et al. ,3HHH-
Marketing Marketing o$ ne) products Spiliopoulos ,<=>7-
Marketing o$ agricultural products Eaourakis et al. ,<==@-
Siskos and Matsatsinis ,<==@-
Eaourakis et al. ,<==K-
Matsatsinis et al. ,<===, 3HHH-
Siskos et al. ,3HH<-
Matsatsinis and Siskos ,3HH<-
Matsatsinis and Siskos ,3HH@-
Consu#er behavior Siskos et al. ,<==?a-
Siskos et al. ,<==?b-
Eaourakis et al. ,<==?-
Manouselis and Matsatsinis ,3HH<-
Matsatsinis ,3HH3-
Custo#er satis$action Grigoroudis et al. ,<===-
Mihelis et al. ,3HH<-
Siskos et al. ,3HH<-
DH '&apter ,
Field Scope 6e$erence
Siskos and Grigoroudis ,3HH3-
Grigoroudis et al. ,3HH3-
Grigoroudis and Siskos ,3HH@-
Sandalidou et al. ,3HH@-
Sales strateg% proble#s 6ichard ,<=>@-
Siskos ,<=>K-
Manage#ent
,general-
1ro.ect evaluation Fac'uet*0agrG/e ,<==?-
Eeuthe et al. ,3HHH-
Environ#ental #anage#ent Siskos and Assi#akopoulos ,<=>=-
Sat/inakos et al. ,<==<-
2iakoulaki et al. ,<===-
Fob evaluation Sp%ridakos et al. ,3HHH-
Gon/[le/*Ara%a et al. ,3HH3-
1. CONC3UDING REMAR5S AND .UTURE RESEARCH
"he !"A #ethods presented in this chapter belong to the $a#il% o$
ordinal regression anal%sis #odels ai#ing to assess a value s%ste# as a
#odel o$ the pre$erences o$ the 2M. "his assess#ent is i#ple#ented
through an aggregation*disaggregation process. Lith this process the anal%st
is able to in$er an anal%tical #odel o$ pre$erences, )hich is as consistent as
possible )ith the 2M+ pre$erences. "he acceptance o$ such a pre$erence
#odel is acco#plished through a repetitive interaction bet)een the #odel
and the 2M. "his approach contributes to)ards an alternative reasoning $or
decision*aid.
Future research regarding !"A #ethods ai#s to e&plore $urther the
potentials o$ the pre$erence disaggregation philosoph% )ithin the conte&t o$
#ulticriteria decision*aid. Fac'uet*0agrG/e and Siskos ,3HH<- propose that
potential research develop#ents #a% be $ocused on
a- the in$erence o$ #ore sophisticated aggregation #odels b%
disaggregation, and
b- the e&peri#ental evaluation o$ disaggregation procedures.
Finall%, it )ould be interesting to e&plore the relationship o$ aggregation
and disaggregation procedures in ter#s o$ si#ilarities and(or dissi#ilarities
regarding the evaluation results obtained b% both approaches ,Fac'uet*
0agrG/e and Siskos, 3HH<-. "his )ill enable the identi$ication o$ the reasons
and the conditions under )hich aggregation and disaggregation procedures
)ill lead to di$$erent or the sa#e results.
,. U%/ 1et&o+s D<
RE.ERENCES
Angilella, S., S. Greco, F. 0a#antia and E. Matara//o ,3HH@-. Assessing non*additive utilit%
$or #ulticriteria decision aid, Iuropean Journal of Kperational 8esearc& ,to appear-.
Eana e Costa, C.A. and F.C. Jansnick ,<==D-. MACEE"S an interactive path to)ards the
construction o$ cardinal value $unctions, International %ransactions in Kperations
8esearc&, < ,D-, D>=*?HH.
Eana e Costa, C.A. and F.C. Jansnick ,<==7-. Applications o$ the MACEE"S approach in
the $ra#e)ork o$ an additive aggregation #odel, Journal of 1ulti7'riteria Decision
/nalysis, K ,3-, <H7*<<D.
Eana e Costa, C.A., F. Nunes 2a Silva and F. C. Jansnick ,3HH<-. Con$lict dissolution in the
public sector A case*stud%, Iuropean Journal of Kperational 8esearc&, <@H ,3-, @>>*DH<.
Eaourakis, G., N.F. Matsatsinis and Y. Siskos ,<==@-. Agricultural product design and
develop#ent, in F. Fanssen and C.S. Skiadas, ,eds.-, /pplie+ stoc&astic mo+els an+ +ata
analysis, Lorld Scienti$ic, <<H>*<<3>.
Eaourakis, G., N.F. Matsatsinis and Y. Siskos ,<==?-. Consu#er behavioural anal%sis using
#ulticriteria #ethods, in F. Fanssen, C.S. Skiadas and C. Vopounidis, ,eds.-, /+vances in
stoc&astic mo+ellin$ an+ +ata analysis, 4lu)er Acade#ic 1ublishers, 2ordrecht, @3>*@@>.
Eaourakis, G., N.F. Matsatsinis and Y. Siskos ,<==K-. Agricultural product develop#ent
using #ultidi#ensional and #ulticriteria anal%ses "he case o$ )ine, Iuropean Journal of
Kperational 8esearc&, =D ,3-, @3<*@@D.
Eeuthe, M. and G. Scannella ,<==K-. Applications co#parNes des #Nthodes d+anal%se
#ulticritGre !"A, 8/I8K 8ec&erc&e KpLrationelle, @H ,@-, 3=@*@<?.
Eeuthe, M. and G. Scannella ,<===-. 1US%/8D userMs $ui+e, G"M, FacultNs !niversitaires
Catholi'ues de Mons ,F!CaM-, Mons.
Eeuthe, M. and G. Scannella ,3HH<-. Co#parative anal%sis o$ !"A #ulticriteria #ethods,
Iuropean Journal of Kperational 8esearc&, <@H ,3-, 3DK93K3.
Eeuthe, M., 0. Eeckhoudt and G. Scanella ,3HHH-. A practical #ulticriteria #ethodolog% $or
assessing risk% public invest#ents, Socio7Iconomic Plannin$ Sciences, @D ,3-, <3<*<@=.
Eui, ".R. ,<=>7-. 'o7oPC / $roup +ecision support system for cooperative multiple criteria
$roup +ecision ma"in$, 0ecture Notes in Co#puter Science, No. 3=H, Springer*Jerlag,
Eerlin.
Charnes, A. and L. Cooper ,<=K<-. 1ana$ement mo+els an+ in+ustrial applications of linear
pro$rammin$ Nol. 1, Lile%, Ne) York.
Charnes, A., L. Cooper, and 6.5. Ferguson ,<=??-. 5pti#al esti#ation o$ e&ecutive
co#pensation b% linear progra##ing, 1ana$ement Science, < ,3-, <@>*<?<.
Cosset, F.C., Y. Siskos and C. Vopounidis ,<==3-. Evaluating countr% risk A decision support
approach, !lo)al 6inance Journal, @ ,<-, 7=*=?.
2espotis, 2.4. and C. Vopounidis ,<==@-. Euilding additive utilities in the presence o$ non*
#onotonic pre$erence, in 1.M. 1ardalos, Y. Siskos and C. Vopounidis ,eds.-, /+vances in
multicriteria analysis, 4lu)er Acade#ic 1ublisher, 2ordrecht, <H<*<<D.
2espotis, 2.4. and 2. Yannacopoulos ,<==H-. MNthode d+esti#ation d+utilitNs additives
concaves en progra##ation linNare #ultiob.ecti$s, 8/I8K 8ec&erc&e KpLrationelle, 3D,
@@<*@D=.
2espotis, 2.4., 2. Yannacopoulos and C. Vopounidis ,<==H-. A revie) o$ the !"A
#ulticriteria #ethod and so#e i#prove#ents, 6oun+ations of 'omputin$ an+ Decision
Sciences, <? ,3-, K@97K.
2evaud, F.M., G. Groussaud and E. Fac'uet*0agrG/e ,<=>H-. !"A27S !ne #Nthode de
construction de $onctions dYutilitN additives rendant co#pte de .uge#ents globau&,
Iuropean Oor"in$ !roup on 1ulticriteria Decision /i+, Eochu#.
D3 '&apter ,
2iakoulaki, 2., C. Vopounidis, G. Mavrotas and M. 2ou#pos ,<===-. "he use o$ a pre$erence
disaggregation #ethod in energ% anal%sis and polic% #aking;, Iner$y%&e International
Journal, 3D ,3-, <?7*<KK.
2ias, 0., J. Mousseau, F. Figueira and F. Cli#aco ,3HH3-. An aggregation(disaggregation
approach to obtain robust conclusions )ith E0EC"6E "67, Iuropean Journal of
Kperational 8esearc&, <@> ,3-, @@3*@D>.
2i#itras, A.7., 6. Slo)inski, 6. Sus#aga, and C. Vopounidis ,<===-. Eusiness $ailure
prediction using rough sets, Iuropean Journal of Kperational 8esearc&, <<D ,3-, 3K@*3>H.
2ou#pos, M. and C. Vopounidis ,3HH3-. 1ulticriteria +ecision ai+ classification met&o+s,
4lu)er Acade#ic 1ublishers, 2ordrecht.
2ou#pos, M. and C. Vopounidis ,3HH3-. 5n the develop#ent o$ an outranking relation $or
ordinal classi$ication proble#s An e&peri#ental investigation o$ a ne) #ethodolog%,
Kptimi<ation 1et&o+s an+ SoftBare, <7 ,3-, 3=@*@<7.
Fishburn, 1. ,<=KK-. A note on recent develop#ents in additive utilit% theories $or #ultiple
$actors situations, Kperations 8esearc&, <D, <<D@*<<D>.
Fishburn, 1. ,<=K7-. Methods $or esti#ating additive utilities, 1ana$ement Science, <@, D@?*
D?@.
Freed, N. and G. Glover ,<=><-. Si#ple but po)er$ul goal progra##ing #odels $or
discri#inant proble#s, Iuropean Journal of Kperational 8esearc&, 7, DD*KH.
Geo$$rion, A.M., F.S. 2%er, and A. Feinberg ,<=73-. An interactive approach $or #ulti*
criterion opti#i/ation, )ith an application to the operation o$ an acade#ic depart#ent,
1ana$ement Science, <= ,D-, @?7*@K>.
Gon/[le/*Ara%a, M.C., 0.A.2. 6angel, M.1.E. 0ins and 0.F.A.M. Go#es ,3HH3-. Euilding
the additive utilit% $unctions $or CA2*!F6F evaluation sta$$ criteria, /nnals of Kperations
8esearc&, <<K ,<*D-, 37<*3>>.
Grabisch, M. ,<==K-. "he application o$ $u//% integrals in #ulticriteria decision #aking,
Iuropean Journal of Kperational 8esearc&, >= ,@-, DD?*D?K.
Grigoroudis, E. and Y. Siskos ,3HH3-. 1re$erence disaggregation $or #easuring and anal%sing
custo#er satis$action "he M!SA #ethod, European Fournal o$ 5perational 6esearch, <D@
,<-, <D>*<7H.
Grigoroudis, E. and Y. Siskos ,3HH@-. A surve% o$ custo#er satis$action baro#eters 6esults
$ro# the transportation*co##unications sector, Iuropean Journal of Kperational
8esearc&, <?3 ,3-, @@D*@?@.
Grigoroudis, E. and Y. Siskos ,3HH@-. M!SA A decision support s%ste# $or evaluating and
anal%sing custo#er satis$action, in 4. Margaritis and 7. 1itas ,eds.-, Procee+in$s of t&e F
t&
Pan&ellenic 'onference in Informatics, "hessaloniki, Greece, <<@*<37.
Grigoroudis, E., F. Malandrakis, F. 1olitis and Y. Siskos ,<===-. Custo#er satis$action
#easure#ent An application to the Greek shipping sector, Procee+in$s of t&e 5
t&
Decision
Sciences InstituteMs International 'onference on Inte$ratin$ %ec&nolo$y & Puman
DecisionsC !lo)al Qri+$es into t&e 01st 'entury, Athens, Greece, 3, <@K@*<@K?.
Grigoroudis, E., Siskos Y. and 5. Saurais ,3HHH-. "E05S A custo#er satis$action evaluation
so$t)are, 'omputers an+ Kperations 8esearc&, 37 ,7*>-, 7==*><7.
Grigoroudis, E., Y. 1olitis and Y. Siskos ,3HH3-. Satis$action bench#arking and custo#er
classi$ication An application to the branches o$ a banking organi/ation, International
%ransactions in Kperational 8esearc&, = ,?-, ?==*K<>.
Sat/inakos, 7., 2. Yannacopoulos, C. Faltsetas and C. Viourkas ,<==<-. Application o$ the
M7N56A decision support s%ste# to the evaluation o$ landslide $avourabilit% in Greece,
Iuropean Journal of Kperational 8esearc&, ?H ,<-, KH*7?.
Surson, Ch. and C. Vopounidis ,<==7-. !estion +e portefeuille et analyse multicritRre,
Econo#ica, 1aris.
,. U%/ 1et&o+s D@
Fac'uet*0agrG/e, E. ,<==H-. 7nteractive assess#ent o$ pre$erences using holistic .udg#ents
"he 16EFCA0C s%ste#, in C.A. Eana e Costa ,ed.-, 8ea+in$s in multiple criteria
+ecision ai+, Springer*Jerlag, Eerlin, 33?*3?H.
Fac'uet*0agrG/e, E. ,<==?-. An application o$ the !"A discri#inant #odel $or the evaluation
o$ 6 pro.ects, in 1.M. 1ardalos, Y. Siskos and C. Vopounidis ,eds.-, /+vances in
multicriteria analysis, 4lu)er Acade#ic 1ublishers, 2ordrecht, 3H@93<<.
Fac'uet*0agrG/e, E. and F. Siskos ,<=7>-. !ne #Nthode de construction de $onctions d+ utilitN
additives e&plicatives d+ une prN$Nrence globale, 'a&ier +u (/1S/DI, <K, !niversitN de
1aris*2auphine.
Fac'uet*0agrG/e, E. and M.F. Shakun ,<=>D-. 2ecision support s%ste#s $or se#istructured
bu%ing decisions, Iuropean Journal of Kperational 8esearc&, <K, D>*?K.
Fac'uet*0agrG/e, E. and Y. Siskos ,<=>3-. Assessing a set o$ additive utilit% $unctions $or
#ulticriteria decision #aking "he !"A #ethod, Iuropean Journal of Kperational
8esearc&, <H ,3-, <?<9<KD.
Fac'uet*0agrG/e, E. and Y. Siskos ,3HH<-. 1re$erence disaggregation 3H %ears o$ MC2A
e&perience, Iuropean Journal of Kperational 8esearc&, <@H ,3-, 3@@93D?.
Fac'uet*0agrG/e, E., 6. Me/iani and 6. Slo)inski ,<=>7-. M501 )ith an interactive
assess#ent o$ a piece)ise linear utilit% $unction, Iuropean Journal of Kperational
8esearc&, @< ,@-, @?H*@?7.
Farke, M., M.". Felassi and M.F. Shakun ,<=>7-. ME27A"56 "o)ard a negotiation support
s%ste#, Iuropean Journal of Kperational 8esearc&, @< ,@-, @<D*@@D.
4arst, 5.F. ,<=?>-. 0inear curve $itting using least deviations, Journal of t&e /merican
Statistical /ssociation, ?@, <<>*<@3.
4eene%, 6.0. ,<=>H-. Sittin$ ener$y facilities, Acade#ic 1ress, Ne) York.
4eene%, 6.0. and S. 6ai$$a ,<=7K-. Decisions Bit& multiple o);ectivesC Preferences an+ value
tra+eoffs, Fohn Lile% and Sons, Ne) York.
4elle%, F.E. ,<=?>-. An application o$ linear progra##ing to curve $itting, Journal of
In+ustrial an+ /pplie+ 1at&ematics, K ,<-, <?*33.
4iss, 0.N., F.M. Martel and 6. Nadeau ,<==D-. E0ECCA0C*An interactive so$t)are $or
#odelling the decision #aker+s pre$erences, Decision Support Systems, <3 ,D*?-, @<<*@3K.
4lein, G., S. Mosko)it/, S. Mahesh and A. 6avindran ,<=>?-. Assess#ent o$ #ultiattribute
#easurable value and utilit% $unctions via #athe#atical progra##ing, Decision Sciences,
<K ,@-, @H=*@3D.
4ostko)ski, M. and 6. Slo)inski ,<==K-. !"AQ Application ,v. <.3H-*!serYs #anual,
Document +u (/1S/DI, No. =?, !niversitN de 1aris*2auphine, 1aris.
Malakooti, E. and Y.W. Vhou ,<==D-. Feed$or)ard arti$icial neural net)orks $or solving
discrete #ultiple criteria decision #aking proble#s, 1ana$ement Science, DH ,<<-, <?D3*
<?K<.
Manas, M. and F. Nedo#a ,<=K>-. Finding all vertices o$ a conve& pol%hedron, Sumerical
1at&ematics, <3, 33K*33=.
Manouselis, N. and N.F. Matsatsinis ,3HH<-. 7ntroducing a #ulti*agent, #ulti*criteria
#ethodolog% $or #odeling electronic consu#er+s behavior "he case o$ internet radio, in
M. 4lush and F. Va#bonelli ,eds.-, (ecture notes in artificial intelli$ence7'ooperative
information a$ents, Springer Jerlag, 3<>3, pp. <=H*<=?.
Marichal, F.0. and M. 6oubens ,3HHH-. 2eter#ination o$ )eights o$ interactive criteria $ro# a
re$erence set, Iuropean Journal of Kperational 8esearc&, <3D ,@-, KD<*K?H.
Matsatsinis, N.F. ,3HH3-. Ne) agricultural product develop#ent using data #ining techni'ues
and #ulticriteria #ethods, in Procee+in$s of t&e 1
st
Pellenic /ssociation of Information
an+ 'ommunication %ec&nolo$y in /$riculture* 6oo+ an+ Invironment 9P/I'%/Ts
'onference 0000:, Fune K*7, Athens, Greece.
DD '&apter ,
Matsatsinis, N.F. and A. Sa#aras ,3HH<-. MC2A and pre$erence disaggregation in group
decision support s%ste#s, Iuropean Journal of Kperational 8esearc&, <@H ,3-, D<D*D3=.
Matsatsinis, N.F. and 1. 2elias ,3HH@-. AgentAllocator An agent*based #ulti*criteria
decision support s%ste# $or task allocation, in J. Marik 2. McFarlane and 1.
Jalckenaers, Polonic an+ multi7a$ent systems for manufacturin$, (ecture notes in
computer science, Springer Jerlag ,to appear-.
Matsatsinis, N.F. and Y. Siskos ,<===-. MA64ER An intelligent decision support s%ste# $or
product develop#ent decision, Iuropean Journal of Kperational 8esearc&, <<@ ,3-, @@K*
@?D.
Matsatsinis, N.F. and Y. Siskos ,3HH<-. 27M7"6A An intelligent decision support s%ste# $or
agricultural products develop#ent decisions, in Procee+in$s of t&e 3
r+
Iuropean
'onference of t&e Iuropean 6e+eration for Information %ec&nolo$y in /$riculture* 6oo+
an+ t&e Invironment 9I6I%/ 0001:, Fune <>*3H, Montpellier, France.
Matsatsinis, N.F. and Y. Siskos ,3HH@-. Intelli$ent support systems for mar"etin$ +ecision,
4lu)er Acade#ic 1ublishers, 2ordrecht.
Matsatsinis, N.F., 1. Moraitis, J. 1so#atakis and N. Spanoudakis ,<===-. 7ntelligent so$t)are
agents $or products penetration strateg% selection, in Procee+in$s of 1o+ellin$
/utonomous /$ents in a 1ulti7/$ent Oorl+ 1FFF 91//1/OTFF:, Fune @H*Ful% 3,
Jalencia, Spain.
Matsatsinis, N.F., 1. Moraitis, J. 1so#atakis and N. Spanoudakis ,3HHH-. Multi*agent
architecture $or agricultural products develop#ent, in Procee+in$s of t&e 0
n+
Iuropean
'onference of t&e Iuropean 6e+eration for Information %ec&nolo$y in /$riculture* 6oo+
an+ t&e Invironment 9I6I%/ 0000:, Septe#ber 37*@H, Eonn, Ger#an%.
Matsatsinis, N.F., 1. Moraitis, J. 1so#atakis and N. Spanoudakis ,3HH<-. An agent*based
s%ste# $or products penetration strateg% selction, /pplie+ /rtificial Intelli$ence Journal
,to appear-.
Mihelis, G., E. Grigoroudis, Y. Siskos, Y. 1olitis and Y. Malandrakis ,3HH<-. Custo#er
satis$action #easure#ent in the private bank sector, Iuropean Journal of Kperational
8esearc&, <@H ,3-, @D7*@KH.
Mousseau, J. and 6. Slo)inski ,<==>-. 7n$erring an E0EC"6E*"67 #odel $ro# assign#ent
e&a#ples, Journal of !lo)al Kptimi<ation, <3 ,3-, <?7*<7D.
Mousseau, J., F. Figueira and F.*1h. Nau& ,3HHHa-. !sing assign#ent e&a#ples to in$er
)eights $or E0EC"6E "67 #ethod So#e e&peri#ental results, Iuropean Journal of
Kperational 8esearc&, <@H ,3-, 3K@*37?.
Mousseau, J., 6. Slo)inski and 1. Vielnie)ic/ ,3HHHb-. A user*oriented i#ple#entation o$
the E0EC"6E*"67 #ethod integrating pre$erence elicitation support, 'omputers an+
Kperational 8esearc&, 37 ,7*>-, 7?7*777.
Muro$ushi, ". and M. Sugeno ,<=>=-. An interpretation o$ $u//% #easure and the Cho'uet
integral as an integral )ithrespect to a $u//% #easure, 6u<<y Sets an+ Systems 3= ,3-, 3H<*
337.
5ral, M. and 5. 4ettani ,<=>=-. Modelling the process o$ #ultiattribute choice, Journal of t&e
Kperational 8esearc& Society, DH ,@-, 3><93=<.
5ral, M., 5. 4ettani, F.C. Cosset and M. 2aouas ,<==3-. An esti#ation #odel $or countr% risk
rating, International Journal of 6orecastin$, > ,D-, ?>@*?=@.
1a)lak, V. ,<=>3-. 6ough sets, International Journal of Information an+ 'omputer Sciences,
<< ,?-, @D<*@?K.
Wuinlan, F.6. ,<=>K-. 7nduction o$ decision trees, 1ac&ine (earnin$, < ,<-, ><*<HK.
6ichard, F.0. ,<=>@-. Aide ] la dNcision stratNgi'ue en 1.M.E.;, in E. Fac'uet*0agrG/e and Y.
Siskos ,eds.-, 1Lt&o+e +e +Lcision multicritRre, So##es et "echni'ues, 1aris, <<=*<D3.
6o%, E. ,<=>?-. 1Lt&o+olo$ie 1ulticritRre +M/i+e U la DLcision, Econo#ica, 1aris.
,. U%/ 1et&o+s D?
Sandalidou, E., E. Grigoroudis and Y. Siskos ,3HH@-. 5rganic and conventional olive oil
consu#ers A co#parative anal%sis using custo#er satis$action evaluation approach, in G.
Eaourakis ,ed.-, Marketing trends $or organic $ood in the advent o$ the 3<
st
centur%, Lorld
Scienti$ic ,to appear-.
Shakun, M.F. ,<=>>-. Ivolutionary systems +esi$nC Policyma"in$ un+er comple@ity an+
$roup +ecision support systems, Solden*2a%, San Francisco, CA.
Shakun, M.F. ,<==<-. Airline bu%out Evolutionar% s%ste#s design and proble# restructuring
in group decision and negotiation, 1ana$ement Science, @7 ,<H-, <3=<*<@H@.
Siskos, F. ,<=>H-. Co##ent #odNliser les prN$Nrences au #o%en de $onctions d+utilitN
additives, 8/I8K 8ec&erc&e KpLrationelle, <D, ?@*>3.
Siskos, F. ,<=>3-. A )a% to deal )ith $u//% pre$erences in #ulticriteria decision proble#s,
Iuropean Journal of Kperational 8esearc&, <H ,@-, @<D*@3D.
Siskos, F. ,<=>@-. Anal%se de s%stG#es de dNcision #ulticritGre en univers alNatoire,
6oun+ations of 'ontrol In$ineerin$, <H, ,@*D-, <=@*3<3.
Siskos, F. ,<=>?-. Anal%ses de rNgression et progra##ation linNaire, 8Lvue +e StatistiEue
/ppliEuLe, 3@ ,3-, D<*??.
Siskos, F. and 2. 4. 2espotis ,<=>=-. A 2SS oriented #ethod $or #ultiob.ective linear
progra##ing proble#s, Decision Support Systems, ? ,<-, D7*??.
Siskos, F. and N. Assi#akopoulos ,<=>=-. Multicriteria high)a% planning A case stud%,
1at&ematical an+ 'omputer 1o+ellin$, <3 ,<H*<<-, <DH<*<D<H.
Siskos, F. and N.F. Matsatsinis ,<==@-. A 2SS $or #arket anal%sis and ne) product design,
Journal of Decision Systems, 3 ,<-, @?*KH.
Siskos, F., A. Sp%ridakos and 2. Yannacopoulos ,<==@-. M7N56A A #ulticriteria decision
aiding s%ste# $or discrete alternatives, Journal of Information Science an+ %ec&nolo$y, 3
,3-, <@K*<D=.
Siskos, F., and C. Vopounidis ,<=>7-. "he evaluation criteria o$ the venture capital invest#ent
activit% An interactive assess#ent, Iuropean Journal of Kperational 8esearc&, @< ,@-,
@HD*@<@.
Siskos, Y. ,<=>K-. Evaluating a s%ste# o$ $urniture retail outlets using an interactive ordinal
regression #ethod, Iuropean Journal of Kperational 8esearc&, 3@, <7=*<=@.
Siskos, Y. ,3HH3-. M!S"A62 Multicriteria utilit%*based stochastic aid $or ranking
decisions, Journal of Qe&avioral Decision 1a"in$, <? ,?-, DK<*DK?.
Siskos, Y. and A. Sp%ridakos ,<===-. 7ntelligent #ulticriteria decision support 5vervie) and
perspectives, Iuropean Journal of Kperational 8esearc&, <<@ ,3-, 3@K*3DK.
Siskos, Y. and 2. Yannacopoulos ,<=>?-. !"AS"A6 An ordinal regression #ethod $or
building additive value $unctions, Investi$aVWo Kperacional, ? ,<-, @=9?@.
Siskos, Y., A. Spiridakos and 2. Yannacopoulos ,<===-. !sing arti$icial intelligence and
visual techni'ues into pre$erence disaggregation anal%sis "he M772AS s%ste#, Iuropean
Journal of Kperational 8esearc&, <<@ ,3-, 3@K*3DK.
Siskos, Y., C. Vopounidis and A. 1oulie/os ,<==D-. An integrated 2SS $or $inancing $ir#s b%
an industrial develop#ent bank in Greece, Decision Support Systems, <3 ,3-, <?<*<K>.
Siskos, Y., E. Grigoroudis, C. Vopounidis and 5. Saurais ,<==>-. Measuring custo#er
satis$action using a collective pre$erence disaggregation #odel, Journal of !lo)al
Kptimi<ation, <3 ,3-, <7?*<=?.
Siskos, Y., E. Grigoroudis, N.F. Matsatsinis and G. Eaourakis ,<==?a-. 1re$erence
disaggregation anal%sis in agricultural product consu#er behaviour, in 1.M. 1ardalos, Y.
Siskos and C. Vopounidis ,eds.-, /+vances in multicriteria analysis, 4lu)er Acade#ic
1ublishers, 2ordrecht, <>?*3H3.
Siskos, Y., E. Grigoroudis, N.F. Matsatsinis, G. Eaourakis and F. Negue/ ,<==?b-.
Co#parative behavioural anal%sis o$ European olive oil consu#er, in F. Fanssen, C.S.
DK '&apter ,
Skiadas and C. Vopounidis ,eds.-, /+vances in stoc&astic mo+ellin$ an+ +ata analysis,
4lu)er Acade#ic 1ublishers, 2ordrecht, 3=@*@<H.
Siskos, Y., E. Grigoroudis, Y. 1olitis and Y. Malandrakis ,3HH<-. Custo#er satis$action
evaluation So#e real e&periences, in A. Colorni, M. 1aruccini and E. 6o% ,eds.-, /7
1'D7/C 1ultiple 'riteria Decision /i+in$, European Co##ission Foint 6esearch Centre,
3=7*@<D.
Siskos, Y., N.F. Matsatsinis and G. Eaourakis ,3HH<-. Multicriteria anal%sis in agricultural
#arketing the case o$ French olive oil #arket, Iuropean Journal of Kperational
8esearc&, <@H, ,3-, @<?*@@<.
Siskos, ^., and E. Grigoroudis ,3HH3-. Measuring custo#er satis$action $or various services
using #ulticriteria anal%sis, in 2. Eou%ssou, E. Fac'uet*0agrG/e, 1. 1ern%, 6. S_lo)i`ski,
2. Janderpooten and 1. Jincke, 1., ,eds.-, /i+in$ +ecisions Bit& multiple criteriaC Issays
in &onor of Qernar+ 8oy, 4lu)er Acade#ic 1ublishers, 2ordrecht, D?7*D>3-.
Slo)inski, 6. ,<==?-. 6ough set approach to decision anal%sis, /I I@pert 1a$a<ine, <H ,@-,
<>*3?.
Spiliopoulos, 1. ,<=>7-. /nalyse et simulation +u marc&L pour le lancement +Mum nouveau
pro+uit 8Lalisation +Mun SI/D, "hGse de @e c%cle, !niversirN de 1aris*2auphine, 1aris.
Sp%ridakos, A., Y. Siskos, 2. Yannakopoulos and A. Skouris ,3HHH-. Multicriteria .ob
evaluation $or large organisations, Iuropean Journal of Kperational 8esearc&, <@H ,3-,
@7?*@>7.
Srinivasan, J. and A.2. Shocker ,<=7@-. 0inear progra##ing techni'ues $or
#ultidi#ensional anal%sis o$ pre$erences, Psyc&ometri"a, @> ,@-, @@79@=K.
Sta#, A., M. Sun and M. Saines ,<==K-. Arti$icial neural net)ork representations $or
hierarchical pre$erence structures, 'omputers an+ Kperations 8esearc&, 3@ ,<3-, <<=<*
<3H<.
Ste)art, ".F. ,<=>7-. 1runing o$ decision alternatives in #ultiple criteria decision #aking,
based on the !"A #ethod $or esti#ating utilities, Iuropean Journal of Kperational
8esearc&, 3> ,<-, 7=*>>.
Jan de 1anne, C. ,<=7?-. 1et&o+s for linear an+ Eua+ratic pro$rammin$, North*Solland
1ublishing Co#pan%, A#sterda#.
Lagner, S.M. ,<=?=-. 0inear progra##ing techni'ues $or regression anal%sis, Journal of t&e
/merican Statistical /ssociation, ?D, 3HK*3<3.
Young, F.L., F. 2e 0eeu), and Y. "akane ,<=7K-. 6egression )ith 'ualitative and
'uantitative variables An alternating least s'uares #ethod )ith opti#al scaling $eatures,
P&yc&ometrica, D< ,D-, ?H?*?3=.
Varas, 4. ,3HHH-. 6ough appro&i#ation o$ a pre$erence relation b% a #ulti*attribute stochastic
do#inance $or deter#inistic and stochastic evaluation proble#s, Iuropean Journal of
Kperational 8esearc&, <@H ,3-, @H?*@<D.
Vionts, S. and F. Lallenius ,<=7K-. An interactive progra##ing #ethod $or solving the
#ultiple criteria proble#, 1ana$ement Science, 33 ,K-, K?3*KK@.
Vopounidis C., N.F. Matsatsinis and M. 2ou#pos ,<==K-. 2eveloping a #ulticriteria
kno)ledge*based decision support s%ste# $or the assess#ent o$ corporate per$or#ance
and viabilit% "he F7NEJA s%ste#, 6u<<y Iconomic 8evieB, < ,3-, @?*?@.
Vopounidis, C. ,<=>7-. A #ulticriteria decision*#aking #ethodolog% $or the evaluation o$ the
risk o$ $ailure and an application, 6oun+ations of 'ontrol In$ineerin$, <3 ,<-, D?*K7.
Vopounidis, C. and M. 2ou#pos ,<==7-. A #ulticriteria decision aid #ethodolog% $or the
assess#ent o$ countr% risk, Iuropean 8esearc& on 1ana$ement an+ Qusiness Iconomics,
@ ,@-, <@*@@.
,. U%/ 1et&o+s D7
Vopounidis, C. and M. 2ou#pos ,<==>-. 2eveloping a #ulticriteria decision support s%ste#
$or $inancial classi$ication proble#s "he F7NC0AS s%ste#, Kptimi<ation 1et&o+s an+
SoftBare, > ,@*D-, 377*@HD.
Vopounidis, C. and M. 2ou#pos ,<===-. Eusiness $ailure prediction using !"A27S
#ulticriteria anal%sis, Journal of t&e Kperational 8esearc& Society, ?H ,<<-, <<@>*<<D>.
Vopounidis, C. and M. 2ou#pos ,3HHHa-. Euilding additive utilities $or #ulti9group
hierarchical discri#ination "he MS27S #ethod, Kptimi<ation 1et&o+s an+ SoftBare, <D
,@-, 3<=93DH.
Vopounidis, C. and M. 2ou#pos ,3HHHb-. 7NJES"56 A decision support s%ste# based on
#ultiple criteria $or port$olio selection and co#position, in A. Colorni, M. 1aruccini and
E. 6o% ,eds.-, /71'D7/C 1ultiple 'riteria Decision /i+in$, European Co##ission Foint
6esearch Centre, @7<*@><.
Vopounidis, C. and M. 2ou#pos ,3HHHc- 16EF27S A #ulticriteria decision support s%ste#
$or sorting decision proble#s, 'omputers an+ Kperations 8esearc&, 37 ,7*>-, 77=*7=7.
Vopounidis, C. and M. 2ou#pos ,3HH<-. A pre$erence disaggregation decision support
s%ste# $or $inancial classi$ication proble#s, Iuropean Journal of Kperation 8esearc&,
<@H ,3-, DH3*D<@.
Vopounidis, C., Ch. Surson and M. 2ou#pos ,3HHH-. 8isEue7PaysC Ivaluation +es aspects
economiEues, Sociau& et 1oliti'ues, Econo#ica, 1aris.
Vopounidis, C., M. 2ou#pos and S.S. Vanakis ,<===-. Stock evaluation using a pre$erence
disaggregation #ethodolog%, Decision Sciences, @H ,3-, @<@*@@K.