Knowledge-Based Systems 59 (2014) 48–57

Contents lists available at ScienceDirect

Knowledge-Based Systems
journal homepage: www.elsevier.com/locate/knosys

Fuzzy analytic hierarchy process with interval type-2 fuzzy sets

Cengiz Kahraman a,⇑, Basßar Öztaysßi a, Irem
_ Uçal Sarı a, Ebru Turanoğlu b
a
Industrial Engineering Department, Istanbul Technical University, 34367, Macka, Istanbul, Turkey
b _
Industrial Engineering Department, Izmir University, 35350 Uckuyular, Izmir, Turkey

a r t i c l e i n f o a b s t r a c t

Article history: The membership functions of type-1 fuzzy sets have no uncertainty associated with it. While excessive
Received 15 November 2012 arithmetic operations are needed with type-2 fuzzy sets with respect to type-1’s, type-2 fuzzy sets gen-
Received in revised form 3 January 2014 eralize type-1 fuzzy sets and systems so that more uncertainty for defining membership functions can be
Accepted 1 February 2014
handled. A type-2 fuzzy set lets us incorporate the uncertainty of membership functions into the fuzzy set
Available online 10 February 2014
theory. Some fuzzy multicriteria methods have recently been extended by using type-2 fuzzy sets. Ana-
lytic Hierarchy Process (AHP) is a widely used multicriteria method that can take into account various
Keywords:
and conflicting criteria at the same time. Our objective is to develop an interval type-2 fuzzy AHP method
Type-2 fuzzy sets
Fuzzy AHP
together with a new ranking method for type-2 fuzzy sets. We apply the proposed method to a supplier
Fuzzy numbers selection problem.
Multicriteria Ó 2014 Elsevier B.V. All rights reserved.
Uncertainty

1. Introduction
In the traditional formulation of multicriteria decision making
(MCDM) problems human’s judgments are represented as exact
numbers. However, in many practical cases, the data may be
imprecise, or the decision makers might be unable to assign exact
numerical values to the evaluation. Since some of the evaluation
criteria are subjective and qualitative in nature, it is very difficult
for the decision maker to express the preferences using exact
numerical values [20]. The conventional MCDM approaches tend
to be less effective in dealing with the imprecise or vague nature
of the linguistic assessments [13]. Analytic Hierarchy Process
(AHP) which is one of the most used MCDM approaches is a struc-
tured multicriteria technique for organizing and analyzing com-
plex decisions including many conflicting criteria. In the
literature fuzzy AHP methods based on type-1 fuzzy sets exist.
The fuzzy AHP technique can be viewed as an advanced analytical
method developed from the traditional AHP. Despite the conve-
nience of AHP in handling both quantitative and qualitative criteria
of multicriteria decision making problems based on decision mak-
ers’ judgments, fuzziness and vagueness existing in many decision-
making problems may cause to the imprecise judgments of deci-
sion makers in conventional AHP approaches [7].
In type-1 fuzzy sets, each element has a degree of membership
which is described with a membership function valued in the
⇑ Corresponding author.
E-mail address: kahramanc@itu.edu.tr (C. Kahraman).
interval [0, 1] [21]. The concept of a type-2 fuzzy set was intro-
duced by Zadeh [22] as an extension of the concept of an ordinary
fuzzy set called a type-1 fuzzy set. Such sets are fuzzy sets whose
membership grades themselves are type-1 fuzzy sets; they are
very useful in circumstances where it is difficult to determine an
exact membership function for a fuzzy set; hence, they are useful
for incorporating linguistic uncertainties, e.g., the words that are
used in linguistic knowledge can mean different things to different
people [14]. While the membership functions of type-1 fuzzy sets
are two-dimensional, the membership functions of type-2 fuzzy
sets are three-dimensional. It is the new third-dimension that pro-
vides additional degrees of freedom that make it possible to di-
rectly model uncertainties.
An interval type-2 fuzzy set is a special case of a generalized
type-2 fuzzy set. Since generalized type-2 fuzzy sets require com-
plex and immense computational burdensome operations, the
wide spread application of generalized type-2 fuzzy systems has
not occurred. Interval type-2 fuzzy sets are the most commonly
used type-2 fuzzy sets because of their simplicity and reduced
computational effort with respect to general type-2 fuzzy sets.
For this reason, we used interval type-2 fuzzy sets.
In this paper, an interval type-2 fuzzy AHP method is developed
and presented into the literature for the first time. The linguistic
scale of fuzzy AHP is expressed in a more detailed and flexible
way by interval type-2 fuzzy sets. New defuzzification methods
for both triangular and trapezoidal type-2 fuzzy sets are also incor-
porated into the developed method.

http://dx.doi.org/10.1016/j.knosys.2014.02.001
0950-7051/Ó 2014 Elsevier B.V. All rights reserved.
 C. Kahraman et al. / Knowledge-Based Systems 59 (2014) 48–57 49

The rest of the paper is organized as follows. Section 2 pre- e e

e1  A
e 2 ¼ aU þ aU ; aU þ aU ; aU þ aU ; aU þ aU ;
A 11 21 12 22 13 23 14 24
sents the basics of interval type-2 fuzzy sets. Section 3 gives          
defuzzification methods including two new methods called DTriT e U e U e U
min H1 A 1 ; H1 A 2 ; min H2 A 1 ; H2 A 2 e U
;
and DTraT. Section 4 includes our proposed interval type-2 fuzzy
aL11 þ aL21 ; aL12 þ aL22 ; aL13 þ aL23 ; aL14 þ aL24 ;
AHP method. Section 5 gives an illustrative application of the pro-          
posed method. Finally Section 6 gives the conclusions. min H1 A e L ; H1 A e L ; min H2 A e L ; H2 A
eL ð4Þ
1 2 1 2

2. Interval type-2 fuzzy sets

Definition 2.3. The subtraction operation between the trapezoidal
  
e
e 1 ¼ aU ; aU ; aU ; aU ; H1 A e U ; H2
In this section, some definitions of type-2 fuzzy sets and inter- interval type-2 fuzzy sets A 11 12 13 14 1
val type-2 fuzzy sets are briefly explained [17].       
e

e e U L L L L e L e
A 1 ; a11 ;a12 ;a13 ;a14 ;H1 A 1 ;H2 A 1L e
and A 2 ¼ a21 ;a22 ;aU23 ;
U U
A type-2 fuzzy set Ae in the universe of discourse X can be rep-
        
resented by a type-2 membership function le , shown as follows e U ;H2 A
aU24 ;H1 A 2
e U ; aL ;aL ;aL ;aL ;H1 A
2 21 22 23 24
e L ;H2 A
2
eL
2 is defined
eA
[22]:
as follows [8]:
  
e 
e¼
A ðx;uÞ; le ðx; uÞ  8x 2 X; 8u 2 J x #½0;1; 0 6 le ðx; uÞ 6 1 ; e e
e1  A
e 2 ¼ aU  aU ; aU  aU ; aU  aU ; aU  aU ;
eA eA A 11 24 12 23 13 22 14 21
         
ð1Þ min H1 A e U ; H1 A
e U ; min H2 Ae U ; H2 AeU ;
1 2 1 2

e
e also can aL11  aL24 ; aL12  aL23 ; aL13  aL22 ; aL14  aL21 ;
where Jx denotes an interval [0, 1]. The type-2 fuzzy set A
         
be represented as follows [17]: min H1 A e L ; H1 A e L ; min H2 A e L ; H2 A
eL ð5Þ
1 2 1 2
Z Z ,
e
e¼
A le ðx; uÞ ðx; uÞ; ð2Þ
x2X u2J x eA Definition 2.4. The multiplication operation between the trape-

RR e
e1 ¼
where Jx # [0, 1] and denote union over all admissible x and u. zoidal interval type-2 fuzzy sets A aU11 ; aU12 ; aU13 ; aU14 ; H1
        
e
e e U ; H2 A e U ; aL ; aL ; aL ; aL ; H 1 A e L ; H2 A eL e
e2 ¼
Let A be a type-2 fuzzy set in the universe of discourse X repre- A 1 1 11 12 13 14 1 1 and A
sented by the type-2 membership function le . If all le ðx; uÞ ¼ 1,        
eA eA aU21 ; aU22 ; aU23 ; aU24 ; H1 A e U ; H2 Ae U ; aL ; aL ; aL ; aL ; H1 A e L ; H2
2 2 21 22 23 24 2
e  
e
then A is called an interval type-2 fuzzy set [3]. An interval type- eL
A 2 is defined as follows [8]:
e
e can be regarded as a special case of a type-2 fuzzy
2 fuzzy set A
set, represented as follows [17]: e e
e1  A
e 2 ffi aU  aU ; aU  aU ; aU  aU ; aU  aU ;
A 11 21 12 22 13 23 14 24
Z Z          
e
e¼ e U e U e U
min H1 A 1 ; H1 A 2 ; min H2 A 1 ; H2 A 2 e U
;
A 1=ðx; uÞ; ð3Þ
x2X u2J x
aL11  aL21 ; aL12  aL22 ; aL13  aL23 ; aL14  aL24 ;
         
where Jx # [0, 1]. e L ; H1 A e L ; min H2 A e L ; H2 A
eL
min H1 A 1 2 1 2 ð6Þ
Arithmetic operations with trapezoidal interval type-2 fuzzy
sets are given in the following.
Definition 2.5. The arithmetic operations between the trapezoidal
Definition 2.1. The upper and lower membership functions of an e
  
interval type-2 fuzzy sets A e 1 ¼ aU ; aU ; aU ; aU ; H1 A
e U ; H2
interval type-2 fuzzy set are type-1 membership functions. 11 12 13 14 1
      
A trapezoidal interval type-2 fuzzy set is illustrated a e U L L L L e L e L
A 1 ; a11 ; a12 ; a13 ; a14 ; H1 A 1 ; H2 A 1 and the crisp value k
       
e
e ¼ A eU; A
e L ¼ aU ; aU ; aU ; aU ; H1 A
e U ; H2 Ae U ; aL ; aL ; aL ; aL ;
A i i i i1 i2 i3 i4 i i i1 i2 i3 i4 is defined as follows [8]:
   
H1 A e L ; H2 A eL where e U and A
A e L are type-1 fuzzy sets,     
i i i i e
e 1 ¼ k  aU ; k  aU ; k  aU ; k  aU ; H1 A e U ; H2 A
eU ;
kA 11 12 13 14 1 1
aUi1 ; aUi2 ; aUi3 ; aUi4 ; aLi1 ,
and aLi2 ; aLi3 aLi4
are the references points of the     
  e ; H2 A
e
e
e e U ; denotes the membership L L L L
k  a11 ; k  a12 ; k  a13 ; k  a14 ; H1 A L L
ð7Þ
interval type-2 fuzzy set A i ; Hj A i
1 1

value of the element aUjðjþ1Þ in the upper trapezoidal membership e 1 1

e
A 1 1 1
    
    ¼  aU11 ;  aU12 ;  aU13 ;  aU14 ; H1 Ae U ; H2 A
eU ;
e U ; 1 6 j 6 2; H A e L denotes the membership value of 1 1
function A j
k k k k k
i i     
the element aLjðjþ1Þ in the lower trapezoidal membership function
1 1 1 1 e L ; H2 AeL
 aL11 ;  aL12 ;  aL13 ;  aL14 ; H1 A 1 1 ð8Þ
      k k k k
e L ; 1 6 j 6 2; H1 A
A e U 2 ½0; 1; H2 A
e U 2 ½0; 1; H1 A
e L 2 ½0; 1; H2
i i i i
  where k > 0.
Ae L 2 ½0; 1 and 1 6 i 6 n [8].
i

3. Defuzzification methods for type-2 fuzzy sets

Definition 2.2. The addition operation between the trapezoidal
  
interval type-2 fuzzy sets A
e
e1 ¼ aU11 ; aU12 ; aU13 ; aU14 ; H1 Ae U ; H2 Defuzzification of a type-2 fuzzy set consists of two steps. In the
1
        first step, a type-2 fuzzy set is determined as a type-1 fuzzy set by
e U ; aL ;aL ;aL ;aL ;H1 Ae L ;H2 A eL e
e 2 ¼ aU ;aU ;aU ; using the type reduction process. Then one of the defuzzification
A 1 11 12 13 14 1 1 and A 21 22 23
         methods for ordinary (type-1) fuzzy sets is used to find the equiv-
U e U e U L L L L e L
a24 ;H1 A 2 ;H2 A 2 ; a21 ;a22 ;a23 ;a24 ;H1 A 2 ;H2 A 2 e L
is defined alence of the type-2 fuzzy set [14]. There are a lot of type reduction
as follows [8]: methods proposed in the literature. In the following the most used
50 C. Kahraman et al. / Knowledge-Based Systems 59 (2014) 48–57

type reduction methods are given and then the proposed ranking where w1 and w2 are the coefficients which satisfy the equation
method is presented. ‘‘w1 + w2 = 1’’.

3.1. Centroid of a type-2 fuzzy set 3.3. Ranking values of interval type-2 fuzzy sets
Lee and Chen [16] presented the concept of ranking values of
The centroid Ce ðxÞ of an interval type-2 fuzzy set is the union of  
A e
ei ¼ A eU; A
e L ¼ aU ; aU ;
the centroids of all its nA embedded type-1 fuzzy sets Ae ½ce ðAe Þ [14]. trapezoidal interval type-2 fuzzy sets. If A i i i1 i2
A         
[ e U ; H2 A
e U Þ; aL ; aL ; aL ; aL ; H1 Ae L ; H2 A e L Þ is a
Ce ¼ e . . . ; cr ð AÞg
ceðAe Þ ¼ fcl ð AÞ; e ¼ ½cl ð AÞ;
e cr ð AÞ;
e ð9Þ aUi3 ; aUi4 ; H1 A i i i1 i2 i3 i4 i i
A A
8Ae e
e iÞ
trapezoidal interval type-2 fuzzy set, the ranking value Rankð A
where e
e i is defined as follows [16]:
PN ! of A
e ¼ min c ðAe Þ ¼ i¼1 xi hi
         
cl ð AÞ min PN ð10Þ e e U þ M1 A
e i Þ ¼ M1 A e L þ M2 A
e U þ M2 A
e L þ M3 A
eU
8Ae eA 8h2½l ðxi Þ;l Rankð A i i i i i
i¼1 hi
 ðxi Þ
eA eA
PN !   1        
e ¼ max c ðAe Þ ¼ i¼1 xi hi þ M3 AeL  S1 Ae U þ S1 A
e L þ S2 Ae U þ S2 A
eL
cr ð AÞ eA max PN ð11Þ i
4 i i i i
8Ae 8h2½l ðxi Þ;l
i¼1 hi
 ðxi Þ
eA eA          
e U þ S3 A
þ S3 A e L þ S4 A
e U þ S4 Ae L þ H1 A eU
Karnik and Mendel [14] proposed iterative algorithms to calcu- i i i i i
e and cr ð AÞ.
late cl ð AÞ e The iterative algorithm for cr ð AÞ
e is given as      
e L þ H2 A
þ H1 A e U þ H2 A
eL ð18Þ
follows: i i i
 
where M p A e j denotes the average of the elements aj and
i ip
Step 1. Set h; hi ¼ ðle ðxi Þ þ l  eðxi ÞÞ=2 for i = 1, . . . , N and com-
A A       
PN P j e ¼j j j ej ¼
pute c ¼ cðh1 ; . . . ; hN Þ ¼ i¼1 xi hi = Ni¼1 hi using Eq. (10).
0 aiðpþ1Þ ; Mp A i aip þ aiðpþ1Þ =2 ; 1 6 p 6 3; Sq A i
rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
Step 2. Find k(1 6 k 6 N  1) such that xk 6 c0 6 xk+1. 1
Pqþ1  j 1
Pqþ1 j 2  
e j
Step 3. Set h ¼ le ðxi Þ for i 6 k and h ¼ l  eðxi Þ for i > k and com- 2 k¼q aik  2 k¼q aik ; 1 6 q 6 3; S4 A i denotes the standard
A A  
pute cr(k) using Eq. (12). deviation of the elements aji1 ; aji2 ; aji3 ; aji4 ; S4 Aej ¼
i
Pk PN rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
   
P P 2
i¼1 xi leA ðxi Þ þ i¼kþ1 xi l eA ðxi Þ 1 4 j 1 4 j e denotes the membership value
j
c rk ¼ ð12Þ 4 k¼1 aik  4 k¼1 aik ; Hp A i
Pk PN
i¼1 leA ðxi Þ þ i¼kþ1 l eA ðxi Þ
of the element ajiðpþ1Þ in the trapezoidal membership function
Step 4. Check if cr(k) = c . If yes, stop. cr(k) is the minimum value
0

e If no, go to step 5. e j ; 1 6 p 6 2; j 2 fU; Lg, and 1 6 i 6 n.

A
of cr ð AÞ. i
   
Step 5. Set c0 = cr(k) and go to Step 2. In Eq. (18) the summation of M1 A e U ; M1 A eL ;
i i
             
e can be obtained by using a procedure sim- e U e L e U e L e U
M2 A i ; M2 A i ;M 3 A i ;M 3 A i ;H1 A i ; H1 A i ;H2 A i e L e U
and
The minimum of cl ð AÞ
 
ilar to the one described above. In step 3, we set h ¼ l e ðxi Þ for i 6 k e
A H2 A L
i is called the basic ranking score, where we deduct the
and h ¼ leðxi Þ for i > k and compute cl(k) using Eq. (13) [14].              
A
Pk PN average of S1 A e U ; S1 Ae L ; S2 Ae U ; S2 A
e L ; S3 A
e U ; S3 A e L ; S4 A
eU
i i i i i i i
i¼1 xi l eA ðxi Þ þ i¼kþ1 xi leA ðxi Þ  
c lk ¼ Pk PN ð13Þ and S4 A e L from the basic ranking score to give the dispersive
i
i¼1 l eA ðxi Þ þ i¼kþ1 leA ðxi Þ
interval type-2 fuzzy set a penalty, where 1 6 i 6 n.

3.2. Type reduction indices method 3.4. Chen and Lee’s likelihood based approach

Niewiadomski et al. [18] proposed optimistic, pessimistic, realis- Chen and Lee [9] proposed the following ranking method for
tic and weighted average indices which determine different points eU P A
type-2 fuzzy sets. They first calculate the likelihood of A eU
s t
e
e is by Eq. (19):
of view for the type reduction of interval type-2 fuzzy sets. If A
 
an interval-valued fuzzy set in the universe X, and le ðxÞ and le ðxÞ

P AeU P A
e U ¼ max
A A s t
are its lower and upper membership functions, Eqs. (14)–(17) trans- 0 0P       1 1
4 P2 e U  Hk Ae U ;0
B k¼1 max atk  atk ;0 þ at4  aS1 þ k¼1 max H
U U U U
e
e e
e k At
form A into an ordinary fuzzy set. TRopt ð AÞ determines optimistic B
@1  max @ P4     
s
  ;0C C
A;0A
aU  aU  þ aU  aU þ aU  aU þ P2 Hk A e U  Hk Ae U 
e
e e
e k¼1 tk tk s4 s1 t4 t1 k¼1 t s
type reduction of A; TRpess ð A Þ determines pessimistic type reduc-
e e
e determines realistic type reduction of A,
e TRre ð AÞ e
e and ð19Þ
tion of A;
e
e e
e Then the ranking values for upper and lower membership functions
TRwa ð AÞ determines weighted average type reduction of A.
are given by Eqs. (20) and (21), respectively:
  !
e   1 Xn  
TRopt Ae ¼l
 e ðxÞ; x 2 X ð14Þ eU ¼
Rank A p A eU þ n  1
eU P A ð20Þ
i i k
A nðn  1Þ k¼1 2
 
e
e ¼ l ðxÞ; x 2 X
TRpess A e ð15Þ and
A
!
  l ðxÞ þ l   1 Xn  
e
e ¼ eA
 e ðxÞ
A eL ¼
Rank A p A eL þ n  1
eL P A ð21Þ
TRre A x2X ð16Þ i
nðn  1Þ k¼1 i k
2
2
 
e e
e i can
TRwa Ae ¼ w1 l ðxÞ þ w2 l  e ðxÞ; x 2 X ð17Þ and finally the ranking values of the interval type-2 fuzzy set A
e A A
be calculated by Eq. (22):
 C. Kahraman et al. / Knowledge-Based Systems 59 (2014) 48–57 51

   
e U þ Rank A
Rank A eL The ranking of the obtained results through defuzzification is con-
i i
eiÞ ¼
Rankð A ð22Þ sistent with the expected order of these triangular type-2 fuzzy
2 sets.
Now we extend Eq. (23) for trapezoidal type-2 fuzzy sets
3.5. The proposed ranking methods (DTraT) as in Eq. (24):
h i
ðuU lU ÞþðbU m1U lU ÞþðaU m2U lU Þ
In the following we propose ranking methods for triangular and 4
þlU þ ðuL lL ÞþðbL m1L 4lL ÞþðaL m2L lL Þ þlL
trapezoidal type-2 fuzzy sets. DTraT ¼
2
We adjusted the center of area (COA) method’s Best Nonfuzzy ð24Þ
Performance (BNP) value for defuzzifying and ranking interval
type-2 fuzzy sets. In our approach, we first obtain defuzzified val- where a and b are the maximum membership degrees of the
ues and then rank fuzzy sets with respect to these values. We pro- lower membership function of the considered type-2 fuzzy set;
pose the Defuzzified Triangular Type-2 Fuzzy Set (DTriT) approach uU is the largest possible value of the upper membership function;
as follows: lU is the least possible value of the upper membership function;
h i m1U and m2U are the second and third parameters of the upper
ðuU lU ÞþðmU lU Þ
3
þ lU þ a ðuL lL Þþðm
3
L lL Þ
þ lL membership function; uL is the largest possible value of the
DTriT ¼ ð23Þ lower membership function; lL is the least possible value of the
2
lower membership function; m1L and m2L are the second and third
where a is the maximum membership degree of the lower member-
parameters of the lower membership function, respectively.
ship function of the considered type-2 fuzzy set; uU is the largest
Using Eq. (24), for Fig. 2a–d the defuzzified values are calculated
possible value of the upper membership function; lU is the least
as follows;
possible value of the upper membership function; mU is the most
h i
possible value of the upper membership function; uL is the largest ð5210Þþð0:852210Þþð3210Þ
4
þ 10 þ ð4715Þþð0:42215Þþð0:63215Þ
4
þ 15
possible value of the lower membership function; lL is the least pos- DTraT 2a ¼ ¼ 25:34
h2 i
sible value of the lower membership function; mL is the most pos- ð5210Þþð0:852210Þþð3210Þ
þ 10 þ ð4722Þþð0:42222Þþð0:63222Þ þ 22
4 4
sible value of the lower membership function. DTraT 2b ¼ ¼ 26:21
h2 i
In the following four different cases are examined: ð5210Þþð0:852210Þþð3210Þ
þ 10 þ ð5210Þþð0:42210Þþð0:63210Þ þ 10
4 4
For Fig. 1a–d the defuzzified values are calculated as follows; DTraT 2c ¼ ¼ 25:34
h i h2 i
ð5210Þþð0:852210Þþð3210Þ
ð4810Þþð2210Þ
þ 10 þ 0:7 ð4215Þþð2215Þ
þ 15 4
þ 10 þ ð3215Þþð0:42215Þþð0:63215Þ
4
þ 15
3 3 DTraT 2d ¼ ¼ 23:46
DTriT 1a ¼ ¼ 22:55 2
2
h i
ð4810Þþð2210Þ ð4822Þþð2222Þ
3
þ 10 þ 0:7 3
þ 22 The proposed method gives a ranking of 1b > 1c > 1a > 1d for trian-
DTriT 1b ¼ ¼ 24:07 gular type-2 fuzzy sets whereas Chen and Lee [9] and Lee and Chen
2
h i [16] gives 1b > 1a > 1c > 1d. For the trapezoidal type-2 fuzzy sets,
ð4810Þþð2210Þ ð4810Þþð2210Þ
3
þ 10 þ 0:7 3
þ 10 the proposed method gives 1b > 1a = 1c > 1d whereas the others
DTriT 1c ¼ ¼ 22:67 do 1b > 1a > 1c > 1d. The ranking of fuzzy sets in Fig. 1a and c are
2
h i somewhat controversial (see Table 1). This is an expected result
ð4810Þþð2210Þ ð2215Þþð2215Þ
þ 10 þ 0:7 þ 15
DTriT 1d ¼
3 3
¼ 20:22 since various ranking methods can give controversial results and
2 this problem is still open to discuss.

Fig. 1. Illustrative examples for triangular type 2 fuzzy sets.

52 C. Kahraman et al. / Knowledge-Based Systems 59 (2014) 48–57

Fig. 2. Illustrative examples for trapezoidal type 2 fuzzy sets.

4. The proposed type-2 fuzzy AHP
In this section, for the proposed method to be better understood,
type-1 fuzzy AHP will first be given with its steps and literature
review. Then, the proposed type-2 fuzzy AHP will be presented.
4.1. Type-1 fuzzy AHP
There are several fuzzy AHP applications which use the fuzzy
set theory to incorporate the linguistic variables into these meth-
ods. Laarhoven and Pedrycz [15] proposed the first algorithm in
fuzzy AHP by describing compared fuzzy ratios with triangular fuz-
zy membership functions. The computation steps in fuzzy AHP are
the same as the steps in crisp AHP which is proposed by Saaty [19].
Logarithmic least square method is used to derive fuzzy weights
and fuzzy performance scores. Buckley [3] presented fuzzy priori-
ties of comparison ratios whose membership functions are trape-
zoidal. He also extended Saaty’s AHP method to incorporate
fuzzy comparison ratios. Buckley [3] claims that Laarhoven and
Pedrycz‘s [15] linear equations do not always have a unique solu-
tion and the obtained solutions are not always triangular fuzzy
numbers. To overcome these difficulties, Buckley used the geomet-
ric mean method to derive fuzzy weights and performance scores
[7]. Buckley’s fuzzy AHP method has received no criticism up to
now. For this reason, Buckley’s method has been chosen to apply
interval type-2 fuzzy sets.
Chang [5] developed extent analysis method for the synthetic
extent values of the pairwise comparisons. The extent analysis
method is used to consider the extent of an object to be satisfied
for the goal. On the basis of the fuzzy values for the extent analysis
of each object, a fuzzy synthetic degree value is obtained.
Table 2 indicates the differences among the fuzzy AHP methods
in the literature. It gives advantages and disadvantages, and main
characteristic of each method.
In the following we briefly give the steps of Buckley’s type-1
fuzzy AHP.
AHP is a structured approach to decision making developed
by Saaty [19]. It is a weighted factor-scoring model and has
the ability to detect and incorporate inconsistencies inherent in
the decision making process. Therefore, it has been applied to
a wide variety of decision making problems, including the eval-
uation of alternatives. Sometimes a decision maker’s judgments
are not crisp, and it is relatively difficult for the decision maker
to provide exact numerical values. Therefore most of the evalu-
ation parameters cannot be given as precisely and the evaluation
data of the alternative project’s suitability for various subjective
criteria and the weights of the criteria are usually expressed in
linguistic terms by the decision makers (DMs). In this case, fuzzy
logic that provides a mathematical strength to capture the
uncertainties associated with human cognitive process can be
used [12,1]. The steps of Buckley’s method are given in the
following:

Table 1
The ranking results of the methods.

Related figures DTriT DTraT Chen and Lee’s likelihood based approach (2010) Lee and Chen’s mean & standard deviation based approach (2008)
Fig. 1a 22.55 (3) – 0.256 (2) 134.7 (2)
Fig. 1b 24.07 (1) – 0.285 (1) 141.1 (1)
Fig. 1c 22.67 (2) – 0.255 (3) 132.9 (3)
Fig. 1d 20.22 (4) – 0.202 (4) 129.0 (4)
Fig. 2a – 25.34 (2) 0.255 (2) 156.7 (2)
Fig. 2b – 26.21 (1) 0.272(1) 161.5 (1)
Fig. 2c – 25.34 (2) 0.253 (3) 154.6 (3)
Fig. 2d – 23.46 (4) 0.218 (4) 152,3 (4)
 C. Kahraman et al. / Knowledge-Based Systems 59 (2014) 48–57 53

Table 2
The comparison of different fuzzy AHP methods [4].

Sources The main characteristics of the method Advantages (A) and disadvantages (D)
Van Laarhoven Direct extension of Saaty’s AHP method with triangular fuzzy numbers Lootsma’s (A) The options of multiple experts can be mod-
and Pedrycz logarithmic least square method is used to derive fuzzy weight and fuzzy performance eled in the reciprocal matrix.
[15] scores. (D) There is not always a solution to linear
equations.
(D) The computational requirement is tremen-
dous, even for a small problem.
(D) It allows only triangular fuzzy numbers to be
used.
Buckley [3] Extension of Saaty’s AHP method with trapezoidal fuzzy numbers Uses the geometric mean (A) It is easy to extend to the fuzzy case.
method to derive fuzzy weights and performance scores (A) It guarantees a unique solution to the reci-
procal comparison matrix.
(D) The computational requirement is
tremendous.
Boender et al. [2] Modifies van Laarhoven and Pedrycz’s method Present a more robust approach to the (A) The options of multiple experts can be
normalization of the local priorities modeled.
(D) The computational requirement is
tremendous.
Chang [5] Synthetical degree values Layer simple sequencing Composite total sequencing (A) The computational requirement is relatively
low.
(A) It follows the steps of crisp AHP. It does not
involve additional operations.
(D) It allows only triangular fuzzy numbers to be
used.
Cheng [10] Builds fuzzy standards Represent performance scores by membership functions (A) The computational requirement is
tremendous.
(D) Entropy is used when probability distribu-
tion is known.
(D) The method is based on both probability and
possibility measures.
Zeng et al. [23] Uses arithmetic averaging method to get performance scores. Extension of Saaty’s AHP (A) It follows the steps of crisp AHP.
method with different scales contains triangular, trapezoidal, and crisp numbers. (A) The options of multiple experts can be
modeled.
(A) There is a flexibility of using different scales.
(D) The computational requirement is tremen-
dous when there are much expert’s
judgements.

Step 1. Pairwise comparison matrices for criteria, subcritaria and also consistent [3]. According to the graded mean integra-
alternatives are constructed using linguistic terms. Each tion approach, a triangular fuzzy number can be trans-
element ða~ij Þ of the pairwise comparison matrix A e is a formed into a crisp number by employing the below
fuzzy number corresponding to its linguistic term. The equation:
pairwise comparison matrix is given by;
  l þ 4m þ u
 1 ~12
a ... a ~1n  A¼ ð27Þ
  6
a 1 ~2n 
 ~21 ... a
e¼
A  ð25Þ If the pairwise comparisons are not consistent, experts
 .. .. ...... .. 
 . . ... .  must reevaluate the pairwise comparisons.

a~ ~n2
a ... 1  Step 3. To weigh the criteria and alternatives, the fuzzy geometric
n1
mean for each row of matrices is computed. First the geo-
Assuming these fuzzy numbers are triangular, Eq. (25) can metric means of the first parameters of triangular fuzzy
be rewritten as in Eq. (26): numbers in each row are calculated:
 
 1 ða12l ; a12m ; a12u Þ . . . ða1nl ; a1nm ; a1nu Þ 

 ða ; a ; a Þ
 a1l ¼ ½1  a12l  . . .  a1nl 1=n
 21l 21m 21u 1 . . . ða2nl ; a2nm ; a2nu Þ 
e ¼
A  .. .. ...... ..

 a2l ¼ ½a21l  1  . . .  a2nl 1=n
 . . ... . 
  ...
 ða ; a ; a Þ ða ; a ; a Þ ... 1 
n1l n1m n1u n2l n2m n2u
ail ¼ ½an1l  an2l  . . .  11=n
ð26Þ
Then the geometric means of the second and third param-
For the evaluation procedure, the linguistic terms given in
eters of triangular fuzzy numbers in each row are calcu-
Table 3 are used.
lated, respectively:
Step 2. The consistency of each fuzzy pairwise comparison matrix
is examined. In order to check the consistency of the fuzzy b1m ¼ ½1  b12m  . . .  b1nm 
1=n

pairwise comparison matrices, pairwise comparison val-

1=n
ues are defuzzified by the graded mean integration b2m ¼ ½b21m  1  . . .  b2nm 
approach [6]. Assume A e ¼ ½a~ij  is a fuzzy positive reciprocal ...
matrix and A ¼ ½a ~ij  is its defuzzified positive reciprocal 1=n
matrix. If the result of the comparisons of is consistent,
bim ¼ ½bn1m  bn2m  . . .  1
then it can imply that the result of the comparisons of is and
54 C. Kahraman et al. / Knowledge-Based Systems 59 (2014) 48–57

c1u ¼ ½1  c12u  . . .  c1nu 1=n Step 1. Fuzzy pairwise comparison matrices among all the criteria
1=n in the dimensions of the hierarchy system are constructed.
c2u ¼ ½c21u  1  . . .  c2nu 
The result of the comparisons is constructed as fuzzy pair-
... wise comparison matrices as following;
ciu ¼ ½cn1u  cn2u  . . .  11=n 2 3 2 3
1 ~~
a ... a ~~ 1 ~~
a ... a ~~
12 1n 12 1n
Assume that the sums of the geometric mean values in the 6~ ~~ 7 6 ~ ~~ 7
6a 1 7 6 1= 1 7
6 a21
e ~ 21 ... a 2n 7 ~ ... a 2n 7
row are a1s for lower parameters; a2s for medium parame- e ¼6
A 6 . ¼6 .
6 . .. .. .. 7
7 6 .. .. .. 77
ters; and a3s for upper parameters. Finally ~rij matrix is ob- 4 . . . . 5 4 .. . . . 5
tained by using aij values obtained above: ~~ ~~ ~~ ~~
a n1 a n2 ... 1 1=a n1 1=a 2n ... 1
 
 Criterion 1 Criterion 2 Criterion j  ð31Þ
     
 a1l b1m c1u a1l b1m c1u a1l b1m c1u 
 a ; a ;a ; ; ; ; ...; ; ; 
 3s 2s 1s a3s a2s a1s a3s a2s a1s  where
       
~r ij ¼  a2l ; b2m ; c2u ; a2l b2m c2u
; ; ; ...; a2l b2m c2u 
; ; 1 1 1 1
 a3s a2s a1s a3s a2s a2s a3s a2s a1s  ~~ ¼
1=a ; U ; U ; U ; H1 aU12 ; H2 aU13 ;
U
  a a a a
       14 13 12 11 
 ail bim ciu
 a3s ; a2s ; a1s ; ail bim ciu
; ; ; ...; ail bim ciu 
; ; 1 1 1 1
a3s a2s a3s a3s a2s a1s  L
; L ; L ; L ; H1 aL22 ; H2 aL23
a24 a23 a22 a21
ð28Þ
The linguistic variables and their triangular and trapezoidal
Step 4. The fuzzy weights and fuzzy performance scores are interval type-2 fuzzy scales which can be used in interval
aggregated as follows: type-2 fuzzy AHP are given in Table 4.
X
n Step 2. The consistency of each fuzzy pairwise comparison matrix
ei ¼
U w
~ j~r ij ; 8i: ð29Þ is examined. Assume A ¼ ½aij  is a positive reciprocal
e
e ¼ ½a
~
j¼1 matrix and A ~ij  is a fuzzy positive reciprocal matrix.
where U e i is the fuzzy utility of alternative i; w ~ j is the If the result of the comparisons A ¼ ½aij  of is consistent,
weight of the criterion j, and ~~r ij is the performance score then it can imply that the result of the comparisons of
e
e ¼ ½a~
~ij  is also consistent. In order to check the consis-
of alternative i with respect to criterion j. A
Step 5. Fuzzy numbers are defuzzified in order to determine the tency of the fuzzy pairwise comparison matrices, the pro-
importance ranking of the criteria and alternatives. The posed DTriT or DTraT approach is used.
Center of Area (COA or Center Index, CI) method can be Step 3. The geometric mean of each row is calculated and then the
used for defuzzification in this step [11]. fuzzy weights are computed by normalization.
The COA method for a triangular fuzzy number can be cal- The geometric mean of each row ~ ~r i is calculated as follows;
culated as follows; ~~ 1=n
~~  . . .  a
~~r i ¼ ½a i1 in
ðui  li Þ þ ðmi  li Þ
BNP i ¼ þ li ; 8i ð30Þ where
3 qffiffiffiffiffi qffiffiffiffiffiffiffi qffiffiffiffiffiffiffi qffiffiffiffiffiffiffi qffiffiffiffiffiffiffi 
where BNP means best nonfuzzy performance.
n
~~ ¼
a n
aUij1 ; n aUij2 ; n aUij3 ; n aUij4 ; Hu1 ðaij Þ; Hu2 ðaij Þ ;
ij
Step 6. The best alternative is determined as in the classical AHP. qffiffiffiffiffiffiffi qffiffiffiffiffiffiffi qffiffiffiffiffiffiffi qffiffiffiffiffiffiffi 
n
aLij1 ; n aLij2 ; n aLij3 ; n aLij4 ; HL1 ðaij Þ; HL2 ðaaij Þ ð32Þ
4.2. Interval type-2 fuzzy AHP

In this section, Buckley’s type-1 fuzzy AHP method will be

modified by using interval type-2 fuzzy sets. The procedure of this
fuzzy AHP method is explained as follows:

Table 3
Linguistic scale for weight matrix.

Linguistic scales Scale of fuzzy number

(1, 1, 3) Equally important
(1, 3, 5) Weakly important
(3, 5, 7) Essentially important
(5, 7, 9) Very strongly important
(7, 9, 9) Absolutely important
Fig. 3. The hierarchy of supplier selection problem.

Table 4
Definition and interval type 2 fuzzy scales of the linguistic variables.

Linguistic variables Triangular interval type-2 fuzzy scales Trapezoidal interval type-2 fuzzy scales
Absolutely Strong (AS) (7.5,9,10.5;1) (8.5,9,9.5;0.9) (7,8,9,9;1,1) (7.2,8.2,8.8,9;0.8,0.8)
Very Strong (VS) (5.5,7,8.5;1)(6.5,7,7.5;0.9) (5,6,8,9;1,1) (5.2,6.2,7.8,8.8;0.8,0.8)
Fairly Strong (FS) (3.5,5,6.5;1)(4.5,5,5.5;0.9) (3,4,6,7;1,1) (3.2,4.2,5.8,6.8;0.8,0.8)
Slightly Strong (SS) (1.5,3,4.5;1)(2.5,3,3.5;0.9) (1,2,4,5;1,1) (1.2,2.2,3.8,4.8;0.8,0.8)
Exactly Equal (E) (1,1,1;1)(1,1,1;1) (1,1,1,1;1,1) (1,1,1,1;1,1)
If factor i has one of the above linguistic variables assigned to it Reciprocals of above Reciprocals of above
when compared with factor j, then j has the reciprocal value
when compared with i
 C. Kahraman et al. / Knowledge-Based Systems 59 (2014) 48–57 55

The fuzzy weight of the ith criterion is calculated as Step 4. The fuzzy weights and fuzzy performance scores are
follows; aggregated as follows:
~~ i ¼ ~~r  ½~~r  . . .  ~~r i  . . .  ~~r n  1 X
n
w ð33Þ ei ¼
U ~~ j~~rij ;
w 8i: ð34Þ
j¼1
where
  e
e i is the fuzzy utility of alternative i; w ~
~~
a au1 au au au where U ~ j is the
~~
ij
; b2u ; b3u ; b4u ; min Hu1 ðaÞ; Hu1 ðbÞ ; min Hu2 ðaÞ; Hu2 ðbÞ
¼ bu4 weight of the criterion j, and ~~r ij is the performance score
b
ij 3 2 1
L L L L     of alternative i with respect to criterion j.
a1 a2 a3 a4
; ; ; ; min HL1 ðaÞ; HL1 ðbÞ ; min HL2 ðaÞ; HL2 ðbÞ
bL bL bL bL Step 5. The classical AHP method’s procedure is applied to deter-
4 3 2 1

mine the best alternative.

Table 5 5. An application
Pairwise comparison matrix for criteria.

P Q D C Consider a supplier selection problem with four criteria and two

P E 1/FS FS VS alternatives (S1 and S2). The hierarchy of the problem is given in
Q FS E VS VS Fig. 3. The considered criteria are price (P), quality (Q), delivery
D 1/FS 1/VS E SS (D), and capacity (C). The abbreviations of the linguistic variables
C 1/VS 1/VS 1/SS E
are represented in Table 3.
We apply the steps of the proposed type-2 fuzzy AHP method in
the following.

Table 6
Step 1. Construction of the pairwise comparison matrices. Tables
Pairwise comparison matrix for alternatives.
5 and 6 present pairwise comparison matrices for criteria
w.r.t. Q w.r.t. P and alternatives using linguistic terms.
S1 S2 S1 S2
Step 2. Consistency check of defuzzified pairwise comparison
S1 E 1/FS S1 E SS matrices. Using the proposed DTraT method, the defuzz-
S2 FS E S2 1/SS E
ified values in Tables 7 and 8 are obtained.
w.r.t.D w.r.t.C The defuzzified pairwise comparison matrix for criteria
S1 S2 S1 S2
has been checked for its consistency and it has been found
S1 E SS S1 E 1/A that its consistency ratio is under 0.1.
S2 1/SS E S2 A E
Step 3. Geometric mean of each row of defuzzified pairwise com-
parison matrices is calculated using Eq. (27). For the repre-
sentative calculations, Table 9 is used, which includes the
type-2 fuzzy sets of pairwise comparisons for the criteria.
Table 7
Defuzzified pairwise comparison matrix for criteria.
For example the geometric mean of the first row is calcu-
lated as given below;
w.r.t. the Goal P Q D C h i1
~~r1 ¼ a~~  a
~~  a ~~ 4
~~  a
P 1 0.2 4.75 6.65 11 12 13 14
Q 4.75 1 6.65 6.65 ¼ ½ð1; 1; 1; 1; 1; 1Þð1; 1; 1; 1; 1; 1Þ  ð0:14; 0:16; 0:25;0:33; 1; 1Þ
D 0.21 0.1 1 2.85
ð0:14; 0:17;0:23; 0:31; 0:8; 0:8Þ  ð3; 4; 6:7; 1;1Þ
C 0.14 0.1 0.44 1
 ð3; 2;4:2; 5:8; 6:8; 0:8;0:8Þ  ð5; 6; 8; 9; 1; 1Þ
1
 ð5:2;6:2; 7:8; 8:8; 0:8;0:8Þ4
1
Table 8 ¼ ½ð2:14; 4:00;12; 21; 1; 1Þð2:44; 4:48; 10:77; 18:7; 0:8; 0:8Þ4
Defuzzified pairwise comparison matrices for alternatives. ¼ ð1:20; 1:41; 1:86; 2:14; 1;1Þ ð1:25; 1:45; 1:81;2:07; 0:8; 0:8Þ
w.r.t. Q w.r.t. P Step 4. The priority weights of criteria and alternatives are deter-
S1 S2 S1 S2
mined by using Eq. (33). For example the priority weights
S1 1 0.21 S1 1 2.85 of the criteria (Table 5) can be calculated as follows.
S2 4.75 1 S2 0.44 1
The calculated geometric means are found in Step 3 as fol-
w.r.t. D w.r.t. C lows;
S1 S2 S1 S2
Using the data represented in Table 10, the priority weight
S1 1 4.75 S1 1 0.116 of P can be calculated as shown below;
S2 0.21 1 S2 7.83 1

Table 9
Pairwise comparisons for the criteria.

P Q D C
P (1,1,1,1;1,1) (1,1,1,1;1,1) (0.14,0.16,0.25,0.33;1,1) (3,4,6,7;1,1) (3.2,4.2,5.8,6.8;0.8,0.8) (5,6,8,9;1,1)
(0.14,0.17,0.23,0.31;0.8,0.8) (5.2,6.2,7.8,8.8;0.8,0.8)
Q (3,4,6,7;1,1) (3.2,4.2,5.8,6.8;0.8,0.8) (1,1,1,1;1,1) (1,1,1,1;1,1) (5,6,8,9;1,1) (5.2,6.2,7.8,8.8;0.8,0.8) (5,6,8,9;1,1)
(5.2,6.2,7.8,8.8;0.8,0.8)
D (0.14,0.16,0.25,0.33;1,1) (0.11,0.12,0.16,0.2;1,1) (1,1,1,1;1,1) (1,1,1,1;1,1) (1,2,4,5;1,1)
(0.14,0.17,0.23,0.31;0.8,0.8) (0.11,0.12,0.16,0.19;0.8,0.8) (1.2,2.2,3.8,4.8;0.8,0.8)
C (0.11,0.12,0.16,0.2;1,1) (0.11,0.12,0.16,0.2;1,1) (0.2,0.25,0.5,1;1,1) (1,1,1,1;1,1) (1,1,1,1;1,1)
(0.11,0.12,0.16,0.19;0.8,0.8) (0.11,0.12,0.16,0.19;0.8,0.8) (0.20,0.26,0.45,0.83;0.8,0.8)
56 C. Kahraman et al. / Knowledge-Based Systems 59 (2014) 48–57

~~ 1 ¼ ~~r  ½~~r 1  ...  ~~r  ...  ~~rn 1

w μ(x)
i i

¼ ð1:20;1:41;1:86;2:14;1;1Þð1:25;1:45;1:81;2:07;0:8;0:8Þ
 ½ð1:20;1:41;1:86;2:14;1;1Þð1:25;1:45;1:81;2:07;0:8;0:8Þ 1
 ð2:94;3:46;4:42;4:87;1;1Þð3:04;3:56;4:33;4:79;0:8;0:8Þ
0.8
ð0:35;0:45;0:63;0:75;1;1Þð0:37;0:46;0:61;0:73;0:8;0:8Þ
1
ð0:22;0:25;0:34;0:44;1;1Þð0:22;0:25;0:32;0:41;0:8;0:8Þ
¼ ð1:20;1:41;1:86;2:14;1;1Þð1:25;1:45;1:81;2:07;0:8;0:8Þ
1
½ð4:73;5:58;7:27;8:22;1;1Þð4:90;5:74;7:09;8:02;0:8;0:8Þ
¼ ð1:20;1:41;1:86;2:14;1;1Þð1:25;1:45;1:81;2:07;0:8;0:8Þ 0.502
ð0:12;0:13;0:17;0:21;1;1Þð0:12;0:14;0:17;0:20;0:8;0:8Þ
0.10 1.07 1.24 x
¼ ð0:14;0:19;0:33;0:45;1;1Þð0:15;0:20;0:31;0:42;0:8;0:8Þ
Fig. 4. Defuzzified value of S1.
Table 10
Geometric means of pairwise comparison matrix for the criteria.

P (1.20,1.41,1.86,2.14;1,1)(1.25,1.45,1.81,2.07;0.8,0.8) Following the same procedure, the fuzzy weights of the

Q (2.94,3.46,4.42,4.87;1,1)(3.04,3.56,4.33,4.79;0.8,0.8)
D (0.35,0.45,0.63,0.75;1,1)(0.37,0.46,0.61,0.73;0.8,0.8)
C (0.22,0.25,0.34,0.44;1,1)(0.22,0.25,0.32,0.41;0.8,0.8)
Step 5. Selection of the best supplier.
Table 11
Type 2 fuzzy weights of the criteria and alternatives.
Priority weights
With respect to the Goal
P (0.14,0.19,0.33,0.45;1,1)(0.15,0.20,0.31,0.42;0.8,0.8)
Q (0.35,0.47,0.79,1.03;1,1)(0.38,0.50,0.75,0.97;0.8,0.8)
D (4.31,6.21,0.11,0.16;1,1)(4.69,6.61,0.10,0.14;0.8,0.8)
C (2.70,3.43,6.15,9.45;1,1)(2.83,3.61,5.73,8.54;0.8,0.8)
With respect to P
S1 (0.30,0.52,1.04,1.54;1,1)(0.35,0.56,0.97,1.41;0.8,0.8)
S2 (0.13,0.18,0.36,0.69;1,1)(0.14,0.19,0.33,0.58;0.8,0.8)
With respect to Q
S1 (0.11,0.13,0.20,0.27;1,1)(0.12,0.14,0.19,0.25;0.8,0.8)
S2 (0.53,0.67,1.01,1.25;1,1)(0.56,0.70,0.97,1.20;0.8,0.8)
With respect to D
S1 (0.30,0.52,1.04,1.54;1,1)(0.35,0.56,0.97,1.41;0.8,0.8)
S2 (0.13,0.18,0.36,0.69;1,1)(0.14,0.19,0.33,0.58;0.8,0.8)
With respect to C
S1 (0.009,0.01,0.11,0.12;1,1)(0.01,0.10,0.10,0.12;0.8,0.8)
S2 (0.78,0.84,0.94,1.00;1,1)(0.79,0.86,0.92,0.98;0.8,0.8)
other criteria and alternatives are calculated as shown in
Table 11.
The priority weight of each alternative is calculated by
summing up the local weights. The local weights are calcu-
lated by multiplying the local weight of the alternative
with the weights of the related criteria. Using the data in
Table 10 and Definition 2.4, the global weights of the alter-
natives are first calculated as Type-2 fuzzy sets (Table 12)
and then overall weights are calculated (Table 13).
According to the results in Table 13, Supplier 1 is selected
with the largest global weight of 0.6158.
The effectiveness of DTraT method can be shown in Figs. 4
and 5. Defuzzified values are at the center of areas of these
type-2 fuzzy sets.
In order to compare our results with Buckley’s type-1
fuzzy AHP method we solved the same problem using
type-1 fuzzy sets. The obtained results with type-1 fuzzy
sets are as in Table 14.
Both methods give the same ranking results. Type-1 and
type-2 fuzzy AHP let us make pairwise comparisons under
uncertainty. However, type-2 fuzzy AHP enable us to
define the membership functions with a larger flexibility.
This flexibility is reflected to the results with a larger

Table 12
Type 2 fuzzy global weights of the alternatives.

Criteria Q P
Criteria (0.35,0.47,0.79,1.03;1,1)
weights (0.38,0.50,0.75,0.97;0.8,0.8)
Alternatives Local weights
S1 (0.11,0.13,0.20,0.27;1,1)
(0.12,0.14,0.19,0.25;0.8,0.8)
S2 (0.53,0.67,1.01,1.25;1,1)
(0.56,0.70,0.97,1.20;0.8,0.8)
Global weights
S1 (0.004,0.10,0.34,0.69;1,1)
(0.005,0.11,0.30,0.59;0.8,0.8)
S2 (0.002,0.03,0.12,0.31;1,1)
(0.002,0.004,0.10,0.24;0.8,0.8)
(0.14,0.19,0.33,0.45;1,1)
(0.15,0.20,0.31,0.42;0.8,0.8)
Local weights
(0.30,0.52,1.04,1.54;1,1)
(0.35,0.56,0.97,1.41;0.8,0.8)
(0.13,0.18,0.36,0.69;1,1)
(0.14,0.19,0.33,0.58;0.8,0.8)
Global weights
(0.004,0.006,0.16,0.28;1,1)
(0.004,0.007,0.14,0.25;0.8,0.8)
(0.19,0.32,0.80,1.29;1,1)
(0.21,0.35,0.73,1.17;0.8,0.8)
D C
(4.31,6.21,0.11,0.16;1,1)
(4.69,6.61,0.10,0.14;0.8,0.8)
Local weights
(0.30,0.52,1.04,1.54;1,1)
(0.35,0.56,0.97,1.41;0.8,0.8)
(0.13,0.18,0.36,0.69;1,1)
(0.14,0.19,0.33,0.58;0.8,0.8)
Global weights
(0.001,0.003,0.11,0.24;1,1)
(0.001,0.003,0.10,0.21;0.8,0.8)
(0.0005,0.01,0.004,0.11;1,1)
(0.0006,0.001,0.003,0.08;0.8,0.8)
(2.70,3.43,6.15,9.45;1,1)
(2.83,3.61,5.73,8.54;0.8,0.8)
Local weights
(0.009,0.01,0.11,0.12;1,1)
(0.01,0.10,0.10,0.12;0.8,0.8)
(0.78,0.84,0.94,1.00;1,1)
(0.79,0.86,0.92,0.98;0.8,0.8)
Global weights
(0.0003,0.0003,0.0006,0.001;1,1)
(0.002,0.0002,0.0006,0.001;0.8,0.8)
(0.02,0.002,0.005,0.009;1,1)
(0.002,0.03,0.005,0.008;0.8,0.8)

Table 13
Type 2 fuzzy and defuzzified overall weights of the alternatives.

Type 2 fuzzy global weights Defuzzified weights Normalized crisp weights

S1 (0.10,0.20,0.63,1.24;1,1) (0.12,0.22,0.56,1.07;0.8,0.8) 0.502 0.3841
S2 (0.23,0.39,1.03,1.81;1,1) (0.26,0.43,0.93,1.59;0.8,0.8) 0.805 0.6158
 C. Kahraman et al. / Knowledge-Based Systems 59 (2014) 48–57 57

μ(x) sets and the obtained results can be compared with ours. Other
multi attribute decision making techniques such as; ELECTRE, VI-
1 KOR, and MACBETH can be also extended by using interval type-
2 fuzzy sets.
0.8 There are limited number of defuzzification and ranking meth-
ods for type-2 fuzzy sets in the literature. New defuzzification and
ranking methods can be proposed and compared with the existing
ones.

0.805
References
0.23 1.59 1.81 x
[1] S. Birgün, C. Kahraman, K.G. Gülen, Fuzzy productivity measurement in
Fig. 5. Defuzzified value of S2.
production systems, in: Cengiz Kahraman, Mesut Yavuz (Eds.), Production
Engineering and Management Under Fuzziness, Springer-Verlag, Berlin, 2010,
pp. 417–430.
Table 14
[2] C.G.E. Boender, J.G. de Graan, F.A. Lootsma, Multicriteria decision analysis with
Comparison of type1 and type2 fuzzy AHP methods.
fuzzy pairwise comparisons, Fuzzy Sets Syst. 29 (1989) 133–143.
Criteria Type-1 AHP Type-2 AHP [3] J.J. Buckley, Fuzzy hierarchical analysis, Fuzzy Sets Syst. 17 (1985) 233–247.
[4] G. Büyüközkan, C. Kahraman, D. Ruan, A fuzzy multicriteria decision approach
Weights with respect Rank Weights with respect Rank for software development strategy selection, Int. J. Gen. Syst. 33 (April–June)
to the goal to the goal (2004) 259–280.
[5] D.Y. Chang, Applications of the extent analysis method on fuzzy AHP, Euro. J.
P 0.282 2 0.257 2
Operat. Res. 95 (1996) 649–655.
Q 0.418 1 0.607 1 [6] S.H. Chen, C.H. Hsieh, Graded mean integration representation of generalized
D 0.168 3 0.087 3 fuzzy number, J. Chin. Fuzzy Syst. Assoc. 5 (1999) 1–7.
C 0.132 4 0.049 4 [7] S.J. Chen, C.L. Hwang, F.P. Hwang, Fuzzy Multiple Attribute Decision Making:
Weights of the alternatives Weights of the alternatives Methods and Applications, Springer-Verlag, Berlin, 1992.
[8] S-M. Chen, L-W. Lee, Fuzzy multiple attributes group decision-making based
S1 S2 S1 S2 on the interval type-2 TOPSIS method, Expert Syst. Appl. 37 (2010) 2790–
2798.
P 0.590 0.410 0.716 0.284
[9] S-M. Chen, L-W. Lee, Fuzzy multiple attributes group-decision making based
Q 0.338 0.662 0.174 0.826 on the ranking values and the arithmetic operations of interval type-2 fuzzy
D 0.590 0.410 0.716 0.284 sets, Expert Syst. Appl. 37 (2010) 824–833.
C 0.252 0.748 0.109 0.891 [10] C-H. Cheng, Evaluating naval tactical missile systems by fuzzy AHP based on
Overall 0.440 0.560 0.384 0.616 the grade value of membership function, Euro. J. Operat. Res. 96 (2) (1996)
priority 343–350.
[11] T.Y. Hsieh, S.T. Lu, G.H. Tzeng, Fuzzy MCDM approach for planning and design
tenders selection in public office buildings, Int. J. Proj. Manage. 22 (2004) 573–
584.
[12] C. Kahraman, A. Beskese, I._ Kaya, Selection among ERP outsourcing alternatives
distinction. In type-1 fuzzy AHP, the significance of the using a fuzzy multi-criteria decision making methodology, Int. J. Product. Res.
48 (2) (2010) 547–566.
distinction is 0.560– 0.440 = 0.120 while it is 0.616–
[13] C. Kahraman, D. Ruan, I. Doğan, Fuzzy group decision-making for facility
0.384 = 0.232 for type-2 fuzzy AHP. location selection, Inform. Sci. 157 (2003) 135–153.
[14] N.N. Karnik, J.M. Mendel, Operations on type-2 fuzzy sets, Fuzzy Sets Syst. 122
(2001) 327–348.
6. Conclusion
[15] P.J.M. van Laarhoven, W. Pedrycz, A fuzzy extension of Saaty’s priority theory,
Fuzzy Sets Syst. 11 (1983) 229–241.
Type-1 fuzzy sets are somewhat problematic in defining mem- [16] L.W. Lee, S.M. Chen, Fuzzy multiple attributes group decision-making based on
bership functions since it is not possible to model uncertainty and the extension of TOPSIS method and interval type-2 fuzzy sets, in: Proceedings
of the 2008 International Conference on Machine Learning and Cybernetic,
imprecision sufficiently. Type-2 fuzzy sets capture this problem by Kunming, China, 2008, pp. 3260–3265.
incorporating footprint of uncertainty into type-1 fuzzy sets. Fuzzy [17] J.M. Mendel, R.I. John, F.L. Liu, Interval type-2 fuzzy logical systems made
AHP based on interval type-2 fuzzy sets has been developed for the simple, IEEE Trans. Fuzzy Syst. 14 (6) (2006) 808–821.
[18] A. Niewiadomski, J. Ochelska, P.S. Szczepaniak, Interval-valued linguistic
first time in this paper. Linguistic scales have been also developed summaries of databases, Control Cybernet. 35 (2) (2006) 415–443.
to be used in the proposed fuzzy AHP method. Thus a flexible def- [19] T.L. Saaty, The Analytic Hierarchy Process: Planning, Priority Setting, Resource
inition opportunity to decision makers has been provided. Allocation, McGraw-Hill, New York, 1980.
[20] M.L. Tseng, Y.H. Lin, A.S. Chiu, C.Y. Chen, Fuzzy AHP-approach to TQM strategy
In this paper we extended Buckley’s fuzzy AHP approach using evaluation, Indust. Eng. Manage. Syst.: An Int. J. 7 (1) (2008) 34–43.
interval type-2 fuzzy sets. Nonsymmetrical interval type-2 mem- [21] L.A. Zadeh, Fuzzy sets, Inform. Control 8 (3) (1965) 338–353.
bership functions may create significant differences in the results [22] L.A. Zadeh, The concept of a linguistic variable and its application to
approximate reasoning– I, Inform. Sci. 8 (3) (1975) 199–249.
with respect to the ones of type-1 AHP methods.
[23] J. Zeng, A. Min, N.J. Smith, Application of fuzzy based decision making
For further research, other fuzzy AHP methods like Chang’s [5] methodology to construction project risk assessment, Int. J. Proj. Manage. 25
or Laarhoven and Pedrycz’s [15] may be extended by type-2 fuzzy (2007) 589–600.