You are on page 1of 10

Applied Soft Computing 10 (2010) 9099

Contents lists available at ScienceDirect

Applied Soft Computing


journal homepage: www.elsevier.com/locate/asoc

A new approach to fuzzy distance measure and similarity measure between two
generalized fuzzy numbers
Debashree Guha, Debjani Chakraborty *
Department of Mathematics, IIT-Kharagpur, Kharagpur 721 302, WB, India

A R T I C L E I N F O

A B S T R A C T

Article history:
Received 6 September 2007
Received in revised form 3 December 2008
Accepted 25 June 2009
Available online 2 July 2009

In this paper a fuzzy distance measure between two generalized fuzzy numbers is developed. The metric
properties of this distance measure are also studied. The new distance measure is compared with the other
fuzzy distance measures proposed by Voxman [W. Voxman, Some remarks on distances between fuzzy
numbers, Fuzzy Sets and Systems 100 (1998) 353365] and Chakraborty and Chakraborty [C. Chakraborty,
D. Chakraborty, A theoretical development on fuzzy distance measure for fuzzy numbers, Mathematical
and Computer Modelling 43 (2006) 254261] and turned out to be more reasonable. A new similarity
measure is also developed with the help of the fuzzy distance measure. Examples are given to compare this
similarity measure with the other similarity measure previously proposed. A decision making scheme is
proposed using this similarity measure and this scheme is found to be more acceptable than the existing
methods due to the fact that it considers the degrees of condence of the experts opinion.
2009 Elsevier B.V. All rights reserved.

Keywords:
Generalized fuzzy numbers
Fuzzy distance
Fuzzy similarity
Absolute value

1. Introduction
The concept of similarity is fundamentally important in almost
every scientic eld. Similarity measures between fuzzy sets have
gained importance due to the widespread applications in diverse
elds like decision making, pattern recognition, machine learning
and market prediction, etc. Similarity measure between two fuzzy
numbers is related to their commonality, in theories of the
recognition, identication, and categorization of objects, where a
common assumption is that the greater the commonality between
a pair of objects, more similar they are. Similarity and distance
measure between two fuzzy numbers are closely related concept.
So it is possible to express similarity measure and distance
measure between fuzzy numbers by a functional relationship. This
is because one of the oldest and most inuential theoretical
assumptions is that similarity measure is inversely related to
distance measure. Therefore the study about the distances
between fuzzy numbers is very much signicant. Thus the
distances between fuzzy sets and as well as fuzzy numbers have
gained more attention from researchers.
A lot of research has been done to construct the distance
measure between fuzzy sets [16]. Recently some researchers have
focused their attention to compute the distances between fuzzy
numbers [8,11,21,25,26].
In the literature, it has been seen that the distance methods
basically compute crisp distance measure for fuzzy numbers. But
* Corresponding author.
E-mail address: debjani@maths.iitkgp.ernet.in (D. Chakraborty).
1568-4946/$ see front matter 2009 Elsevier B.V. All rights reserved.
doi:10.1016/j.asoc.2009.06.009

as Voxman [22] said if we are not certain about the numbers


themselves how can we be certain about the distances among
them, hence it is not reasonable to dene crisp distance between
fuzzy objects. With this point of view Voxman [22] rst
introduced the concept of fuzzy distance measure between two
normal fuzzy numbers using the concept of a cut and in 2006,
Chakraborty and Chakraborty [1] dened an improved fuzzy
distance measure.
Now here, in this paper, a detail study has been made to show
that the distance measure proposed by Chakraborty and Chakraborty [1] is not effective in some cases. And a new fuzzy distance
measure has been proposed which can overcome the drawbacks of
the distance measure proposed by Chakraborty and Chakraborty
[1]. Also the fuzzy distance measures proposed in the paper [1,22]
compute the distance between two normal fuzzy numbers only. In
this paper, the denition of generalized fuzzy numbers, given by
Chen [3,5] has been considered. And the new fuzzy distance
measure can calculate the distance measure between two fuzzy
numbers with different condence level.
After giving the denition of the distance measure, a new
concept of similarity measure between two generalized fuzzy
numbers has been introduced in this paper. Similarity measures
play a very important role in fuzzy decision making [2,17] and
fuzzy pattern recognition [1820,27]. Some other methods have
been also presented to calculate the degree of similarity [4,7,15,24]
between fuzzy numbers.
In the literature all the existing similarity measures give crisp
value for the similarity between two fuzzy numbers. But in any real
eld; when two experts give their responses in terms of linguistic

D. Guha, D. Chakraborty / Applied Soft Computing 10 (2010) 9099

expression, the similarity between their responses must also be


expressed linguistically (refer Section 4 for a more detail
explanation). This is our motivation to develop a new distance
based similarity measure which will give a fuzzy number as the
similarity between two generalized fuzzy numbers. Introducing
this fuzzy similarity measure, sets of examples have been given to
show that the proposed similarity measure is more effective and
efcient than the existing ones.
In Section 2, the denition of generalized fuzzy numbers [3,5]
and also the denition of absolute value of a fuzzy number [13]
have been given. The fuzzy distance measures proposed by
Voxman [22], Chakraborty and Chakraborty [1] have been
described in this section. The similarity measures introduced by
Chen [4], Lee [17], and Chen and Chen [7] have also been discussed.
The new fuzzy distance measure has been derived in Section 3. The
metric properties have also been studied. Then it has been shown
that the ambiguity of the proposed distance measure is less than
that of Chakraborty and Chakraborty [1] which in turn is less than
that of Voxman. In Section 4, the distance based similarity measure
has been proposed and sets of examples have been given. In
Section 5, the proposed similarity measure of generalized fuzzy
numbers has been used to deal with fuzzy decision making
problem, where the experts condence level is incorporated. A
short conclusion has been given in Section 6.

Ca

8
>
< maxfAL2 a  AR1 a; 0g
>
: maxfAL a  AR a; 0g
1

2. Preliminaries

In the year 1985 and 1999, Chen [3,5] represented a


generalized trapezoidal fuzzy number A as A a1 ; a2 ; b; g ; w,
where 0 < w  1, and a1, a2, b and g are real numbers, w
The
represents the degree of condence of expert regarding A.
generalized fuzzy number A is a fuzzy subset on the real line
R, whose membership function mA satises the following
conditions:

mA is a continuous mapping from R to the closed interval [0, 1];


mA x 0, where-1 < x  a1  b;
mA x is strictly increasing on [a1  b, a1];
mA x w, where a1  x  a2;
mA x is strictly decreasing on [a2, a2 + g];
mA x 0, where a2 + g  x < 1.

If w 1, then the generalized fuzzy number A is called a normal


trapezoidal fuzzy number and denoted as A a1 ; a2 ; b; g . If b = 0
and g = 0, then A is called a crisp interval. If, a1 = a2 then A is called a
generalized triangular fuzzy number. If b = 0, g = 0 and a1 = a2 and
w 1, then A is called a real number.
2.2. The absolute value of fuzzy number
Let A 2 E1 , where E1 denotes the class of all fuzzy numbers. The
and dened
absolute value of the fuzzy number A is denoted by jAj
[13] as follows:

jAxj

0;

Ax
_ Ax;

8
>
AL a  0
< Aa ;
L
R

and for all a 2 0; 1; jAja 0; jA aj _ A a; AL a  0  AR a


>
:
AR a; AL a; AL a  AR a  0
a AL a; AR a is the a cut representation of A and jAj
a
where A
respectively.
is the a cut representation of jAj,
2.3. Note on the fuzzy distance measure and similarity measure
2.3.1. Fuzzy distance measure between fuzzy numbers
Though several distance measures [8,11,21,25,26] have already
been introduced for computing crisp distances for fuzzy numbers,
it is intuitively more reasonable to conclude that the distance
between two uncertain numbers should also be an uncertain
number.
Now we briey describe the fuzzy distance measure proposed
by Voxman [22]. Let us consider two normalized fuzzy numbers A 1
and A 2 , denoted as A 1 a1 ; a2 ; b1 ; g 1 and A 2 a3 ; a4 ; b2 ; g 2 . The
a-cut of A 1 and A 2 represent following two intervals respectively
A 1 a AL1 a; AR1 aa; A 2 a AL2 a; AR2 a for all a 2 [0,1].
Then the a cut representation of the distance measure [22]
proposed by Voxman say dVoxman is denoted by [C(a), D(a)] and is
dened in the following way:
1 L
1
A a AR1 a  AL2 a AR2 a
2 1
2
1
1
if AL2 a AR2 a  AL1 a AR1 a
2
2
if

and

2.1. Basic concepts of generalized fuzzy numbers

(1)
(2)
(3)
(4)
(5)
(6)

91

x<0
x0

Da maxfAR1 a  AL2 a; AR2 a  AL1 ag


In 2006, Chakraborty and Chakraborty [1] proposed the
following fuzzy distance measure, which computes fuzzy distance
between two normalized fuzzy numbers, and it has been shown
that fuzziness and ambiguity of this distance measure is less than
that of Voxman. In this work the fuzzy distance between two
normalized fuzzy numbers A 1 and A 2 is dened by
A 1 ; A 2 dL ; dR ; u; s
d
a1 a1
nR
o
R1
1
L
where u da1  max 0 dLa da; 0 and s 0 dRa da  dRa1

(1)

Here dLa and dRa are dened as follows:


L

da hAL1 a  AL2 a AR1 a  AR2 a AL2 a  AR1 a


dRa hAL1 a  AL2 a AR1 a  AR2 a AR2 a  AL1 a
8
>
AL 1 AR1 1 AL2 1 AR2 1
>
< 1 if 1

2
2
for h
L
R
>
A 1 A1 1 AL2 1 AR2 1
>
: 0 if 1
<
2
2
2.3.2. Example
Here we have considered an example to show that the distance
measure, given in Denition 2.3.1, proposed by Chakraborty and
Chakraborty [1], is not able to calculate the distance properly for
some cases.
Let A 0; 0; 0; 0 and B 0; 0; 0; 0:33 be two normalized
fuzzy numbers. Then from Eq. (1), the distance measure between A
and B is d 0; 0; 0; 0. It is not a satisfactory result. Since A and B
are not identical, therefore their distance measure must not be
zero. Therefore, this result is far from reality and there is a need to
develop more general fuzzy distance measure.

92

D. Guha, D. Chakraborty / Applied Soft Computing 10 (2010) 9099

2.3.3. Similarity measures for fuzzy number


In this section, we briey describe some existing similarity
measures of fuzzy numbers.
Different authors have constructed different similarity measures between two normal trapezoidal fuzzy numbers
A 1 a1 ; a2 ; b1 ; g 1 ; A 2 a3 ; a4 ; b2 ; g 2 . Latest two of them are
discussed here.
In 1996, Chen proposed a similarity measure [4] as follows:
ja1  b1  a3  b2 j ja1  a3 j ja2  a4 j
ja2 g 1  a4 g 2 j
sA 1 ; A 2 1 
4
where sA 1 ; A 2 2 0; 1. Where jaj denotes the absolute value of the
real number a.
In [17], Lee proposed another similarity measure for two
normal trapezoidal fuzzy numbers given below:




A1  A 2 
lp

 41= p
sA1 ; A2 1 
kU k
where U is the universe of discourse.




p
p
A1  A 2  ja1  b1  a3  b2 j ja1  a3 j

Here, sA and sA are the lengths of the bases of the generalized


1
2
trapezoidal fuzzy numbers A 1 and A 2 , respectively, and are dened
as follows:
sA 1 a2 g 1  a1  b1

and

sA 2 a4 g 2  a3  b2 :

The value of BsA ; sA is used to determine whether we


1
2
consider COG based distance or not. If the generalized fuzzy
numbers A 1 and R j are real numbers (i.e. sA sA 0), then the
1
2
COG distance is not considered (i.e. BsA ; sA 0). If either A 1 or A 2
1
2
is a generalized fuzzy number or A 1 and A 2 are generalized fuzzy
numbers (i.e. sA sA 6 0), then COG based distance must
1
2
considered (i.e. BsA ; sA 1).
1
2
The similarity measure proposed by Chen and Chen [7],
calculates the similarity measure of generalized fuzzy numbers
in most situations correctly. But, a situation may arise that two
different generalized fuzzy numbers have the same center-ofgravity (COG). For such cases, the above measure fails (please refer
Section 4.3 for detail).
Further all the existing similarity measures give crisp value. So
there is a necessity to develop a fuzzy similarity measure between
two generalized fuzzy numbers which overcome the drawbacks of
the existing similarity measures.

lp

3. Fuzzy distance measure for generalized fuzzy number

1= p

ja2  a4 j p ja2 g 1  a4 g 2 j p

3.1. Construction of the fuzzy distance measure


and jjUjj=max(U)  min(U)
The methods mentioned above calculate the similarity measure
between normal trapezoidal fuzzy numbers and the condence
level of the fuzzy numbers is not considered. Therefore in 2003,
Chen and Chen proposed a similarity measure [7] between two
generalized trapezoidal fuzzy numbers based on the concept of
center-of-gravity (COG). If A 1 a1 ; a2 ; b1 ; g 1 ; w1 is a generalized
trapezoidal fuzzy number, then the COG point xA ; yA of the
1

generalized fuzzy number A 1 is calculated as follows:

yA



8
a2  a1
>
>
w

2
> 1
<
a2 g 1  a1 b1
;

6
>
>
>
: w1
2

Let us consider two generalized trapezoidal fuzzy numbers as


follows:
A 1 a1 ; a2 ; b1 ; g 1 ; w1 and A 2 a3 ; a4 ; b2 ; g 2 ; w2

a cut representation of A 1 is denoted


for 0 < a  w1 .
a cut representation of A 2 is denoted
for 0 < a  w2 .
Now, the distance between A 1 a and
one of the following

2w1

a
:

6
6
sA 1 ; A 2 61 
4

ja1  b1  a3  b2 j ja1  a3 j

A1

BsA1 ; sA 2

AL1 w1 AR1 w1 AL2 w2 AR2 w2



2
2

A 2 a  A 1 a

if

AL1 w1 AR1 w1 AL2 w2 AR2 w2


<
2
2

In order to consider both the notations together, an indicator


variable h is introduced such that

for

A2

Here the COG points COG A 1 and COG A 2 of A 1 and A 2 are


denoted as xA ; yA and xA ; yA , respectively.
1 sA 1 sA 2 > 0
0 sA 1 sA 2 0

if

hA 1 a  A 2 a 1  hA 2 a  A 1 a La; Ra

ja2  a4 j ja2 g 1  a4 g 2 j7
7
7
5
4

iBS ;S miny ; y
h
A1 A2


A1
A2

 1  xA  xA2 
1
maxy ; y

A 1 a  A 2 a

or

Then the degree of similarity sA 1 ; A 2 between generalized


fuzzy numbers A 1 a1 ; a2 ; b1 ; g 1 ; w1 and A 2 a3 ; a4 ; b2 ; g 2 ; w2
is calculated in the following way:
2

A 2 a for all a 2 0; 1 is

if a1  b1 a2 g 1 and 0 < w1  1

yA a1 a2 a2 g 1 a1 b1 w1  yA

by A 2 a AL2 a; AR2 a

if a1  b1 6 a2 g 1 and 0 < w1  1

and
xA

by A 1 a AL1 a; AR1 a

8
>
>
<1
>
>
:0

AL1 w1 AR1 w1 AL2 w2 AR2 w2



2
2
L
R
A w1 A1 w1 AL2 w2 AR2 w2
<
if 1
2
2
if

(2)

where La hAL1 a  AL2 a AR1 a  AR2 a AL2 a  AR1 a


and Ra hAL1 a  AL2 a AR1 a  AR2 a AR2 a  AL1 a
The a cut representation of the distance measure between two
fuzzy numbers A 1 and A 2 is denoted by dLa ; dRa  for a2[0,w]. By
applying Zadehs extension principle [12], it can be said that

D. Guha, D. Chakraborty / Applied Soft Computing 10 (2010) 9099

w minw1 ; w2 . Now dLa ; dRa  is dened as follows:



La  0
La; Ra;
for all a 2 0; w
dLa ; dRa 
0; jLaj _ Ra; La  0  Ra

From Eq. (3); we have the following fuzzy distances for


(
dLa AL2 a  AR1 a or 0
(I.a) A 1 and A 2 :
dRa AR2 a  AL1 a
8
< d L AL a  AR a or 0
3
2
a
(I.b) A 2 and A 3 :
: R
R
d A3 a  AL2 a
8_ a
< dL AL a  AR a or 0
3
1
a
(I.c) A 1 and A 2 :
:_R
da AR3 a  AL1 a
Therefore we have to prove that

(3)
Therefore the fuzzy distance between A 1 and A 2 is dened by
A 1 ; A 2 dL ; dR ; u; s
d
aw aw

(4)

where u and s is dened in the following way:


Z w

u dLaw  max
dLa da; 0
and s
0
Z w




dRa da  dRaw  for; w minw1 ; w2
0

93

_R

_L

d d d d  d d

(5)

From the above three options (I.a), (I.b) and (I.c), the following
eight combinations are possible:

3.2. More on the example of Section 2.3.2

L
_L
(i) dLa AL2 a  AR1 a; d a 0; da AL3 a  AR1 a

As shown in the example given in Section 2.3.2, the distance


measure obtained was dchakraborty 0; 0; 0; 0. But with the help of
Eq. (4), it has been seen that d pro posed 0; 0; 0; 0:33=2, which is
denitely a better result.

_L

(ii) dLa 0; d a AL3 a  AR2 a; da AL3 a  AR1 a


L

_L

(iii) dLa 0; d a 0; da AL3 a  AR1 a


L
_L
(iv) dLa AL2 a  AR1 a; d a 0; da 0

3.3. Metric properties

_L

_L

(v) dLa AL2 a  AR1 a; d a AL3 a  AR2 a; da AL3 a  AR1 a


The new distance measure satises the following properties of a
distance metric.
(i) The fuzzy distance measure proposed in Section 3.1 is a
positive fuzzy number from Eq. (4).
A 1 ; A 2 , for any two generalized fuzzy numbers
A 1 ; A 2 d
(ii) d

A1 and A2 , by Eq. (2).


(iii) For three generalized trapezoidal fuzzy numbers A 1 , A 2 and A 3 ,
with condence level w, the distance measure satises the
triangle inequality

(vii) dLa AL2 a  AR1 a; d a AL3 a  AR2 a; da 0


L

Proof. Proof of properties (i) and (ii) follows from Eqs. (4) and (2),
respectively.
Proof of property (iii) is given here:
Let A 1 , A 2 and A 3 are three generalized fuzzy numbers with their
a cut representation A i a ALi a; ARi a for i=1, 2, 3 for a 2 0; w,
0 < w  1.
Depending on the relative positions of the means of A 1 ; A 2 and
A 3 , three situations arise.
Situation (I):
When mean of A 1  mean of A 2  mean of A 3
AL1 w AR1 w AL2 w AR2 w AL3 w AR3 w


;
2
2
2

_L

Now, from the above eight combinations, (vii) and (viii) are not
possible.
As from (vii), it can be seen that
A 1 and A 2 are disjoint.
A 2 and A 3 are disjoint.
A 1 and A 3 intersect.

Since in this situation, mean of A 1  mean of A 2  mean


of A 3 , it is very clear that the above case (vii) cannot be
happened.
From (viii), we observe that




A 1 and A 2 intersect.
A 2 and A 3 are disjoint.
A 1 and A 3 intersect.

The situation (viii) is also not possible.


Therefore, for the rest six different cases, the proof of inequality
(5) is given below:

(i) dLa AL2 a  AR1 a; d a 0; da AL3 a  AR1 a


R
L
R
L
d d d d
Z w
Z w
Z w
AL2 a  AR1 a da
AR2 a  AL1 a da
AR3 a  AL2 a da
0
0
0
Z w
Z w
Z w

AL2 a  AR1 a da
AR3 a  AL1 a da
AR2 a  AL2 a da
0
0
0
Z w
Z w
Z w

AR2 a  AR1 a da
AR3 a  AL1 a da
AL2 a  AL2 a da
0
0
0
Z w
Z w

AL3 a  AR1 a da
AR3 a  AL1 a da
0
_L

_R

d d

Proved

_L

(viii) dLa 0; d a AL3 a  AR2 a; da 0





A 2 ; A 3  d
A 1 ; A 3
A 1 ; A 2 d
d

L
_L
(vi) dLa 0; d a 0; da 0

94

D. Guha, D. Chakraborty / Applied Soft Computing 10 (2010) 9099

L
_L
(ii) dLa 0; d a AL3 a  AR2 a; da AL3 a  AR1 a
R
L
R
L
d d d d
Z w
Z w
Z w

AR2 a  AL1 a da
AL3 a  AR2 a da
AR3 a  AL2 a da
0
0
0
Z w
Z w
Z w

AL3 a  AL2 a da
AR3 a  AL1 a da
AR2 a  AR2 a da
0
0
0
Z w
Z w

AL3 a  AR1 a da
AR3 a  AL1 a da
0
_L

_R

d d

Proved

L
_L
L
L
R
(iii) da 0; d a 0; da ZA3 a  A1 a
Z w
w
R
L
R
L
d d d d
AR2 a  AL1 a da
AR3 a  AL2 a da
0
0
Z w
Z w

AR2 a  AL2 a da
AR3 a  AL1 a da
Z0 w
Z0 w

AL3 a  AR1 a da
AR3 a  AL1 a da
0
_L

_R

d d

Proved
L

_L

L
L
R
(iv) da A2 a  A1 a;Zd aw 0; da 0
Z w
Z w
R
L
R
L
AL2 a  AR1 a da
AR2 a  AL1 a da
AR3 a  AL2 a da
d d d d
0
0
0
Z w
Z w
Z w

AL2 a  AR1 a da
AR3 a  AL1 a da
AR2 a  AL2 a da
0
0
Z0 w

AR3 a  AL1 a da
0
_L

_R

d d

proved
L

_L

L
L
R
L
R
L
R
(v) da A2 a  A1 a;Zd a A3 a  A2 a; da ZA3 a  A1 a
w
w
R
L
R
L
d d d d
AL2 a  AR1 a da
AL3 a  AR2 a da
0
0
Z w
Z w

AR2 a  AL1 a da
AR3 a  AL2 a da
Z0 w
Z0 w
Z w

AL2 a  AR1 a da
AL3 a  AR2 a da
AR3 a  AL1 a da
0
Z0 w
Z0 w

AL3 a  AR1 a da
AR3 a  AL1 a da
0
_L

_R

d d

proved

_L

L
(vi) da 0; d a 0; da Z0
Z w
w
R
L
R
L
d d d d
AR2 a  AL1 a da
AR3 a  AL2 a da
0
0
Z w

AR3 a  AL1 a da
0
_L

_R

d d

proved

Proceeding in a similar way as for situation (I), the inequality (5)


is proved for the following two:
(II)

AL1 wAR
w
1
2

AL3 wAR
w
3
2

AL2 wAR
w
2
2

(III)

AL2 wAR
w
2
2

AL1 wAR
w
1
2

AL3 wAR
w
3
2

3.4. Notation of ambiguity of a fuzzy number


Delgado [9,10], introduced the concept of ambiguity to realize
the information captured by a fuzzy number. Ambiguity of a fuzzy
number is measured in the following way. Let us consider a fuzzy
which has the a cut representation as AL a; AR a.
number A,
Then ambiguity of A is dened as follows:
Z 1

AmbA
sa AR a  AL a da
(6)
0

where sa is called a reducing function, which is an increasing


function; s : 0; 1 ! 0; 1 with the property that s0 0 and
s1 1. Different a cuts have different inuence over the
measurement of ambiguity of a fuzzy number, reducing function
is considered for weighting the different a cuts.
Now, it is obvious that, for the case of fuzzy distance measure
some vagueness is always presented in the distance measure. But
certainly, in real life decision making situations, the distance
measure with less vagueness is more suitable and acceptable to us
especially from the stability point of view. Suppose we have two
fuzzy numbers A 1 and A 2 with the same central value(s) but with
different spreads. Then A 1 is expected to be better than A 2 in the
sense of stability or preciseness if
(a) AmbA 1 < AmbA 2
Considering this, our interest is now to compare between the
proposed measure, d pro posed , and the measure proposed by

D. Guha, D. Chakraborty / Applied Soft Computing 10 (2010) 9099

Chakraborty and Chakraborty, dchakraborty , with the help of the


following proposition:
Proposition 1. In the consideration of two normal trapezoidal fuzzy
numbers A 1 and A 2 with their a cuts AL1 a; AR1 a and AL2 a; AR2 a,
respectively. The ambiguity of the fuzzy distance obtained from the
proposed measure is less than or equal to that for the distance measure
of Chakraborty and Chakraborty [1] i.e.
Ambd pro posed  Ambdchakraborty

dRa maxfjAL2 a  AR1 aj _ AR2 a  AL1 ag:


Therefore,
Ambd pro posed
if

(7)
Ambd pro posed

Proof. For h = 0,1;


Ambdchakraborty

if

1
0

saAR1 a AR2 a  AL1 a  AL2 a da

(8)

Now, for our proposed distance measure (in Section 3.1), the
following two situations can arise:
(I) L(a)  0 and (II) L(a)  0  R(a) (with the help of Eq. (3),
we can write the above two)
In case of situation (I), for both h = 0 and 1, from the denition of
ambiguity it is very clear that Ambd pro posed Ambdchakraborty .
For situation (II), two different cases arise depending on the
value of h. The proof of (7) for two different cases is given
separately as follows:
(II.a) For h = 1;
The a cut of the proposed distance measure (from Eqs. (2) and
(3)) can be written as follows:
dLa 0

Therefore
Ambd pro posed
if
Ambd pro posed
if

0
AR1

sa AR1 a  AL2 a

a  AL2 a > jAL1 a  AR2 aj

sa

0
AR1

AR2

AL1

AL2

jAL1

a 

a 

a <

(9)

(10)

Now, Eq. (8) is compared with the Eqs. (9) and (10) respectively
as follows:
(i) From Eqs. (8) and (9), we can write
Z 1
Z 1
sa AR1 a  AL2 a da <
sa AR1 a  AL2 a AR2 a
0

AL1

a da
AR2

AL1

(since L(a)  0, therefore


a  a is positive).
Therefore Ambd pro posed < Ambdchakraborty
(ii) From Eqs. (8) and (10), we can write
Z 1
Z 1
sa AR2 a  AL1 a da <
sa AR2 a  AL1 a AR1 a
0

 AL2 a da
(as dRa AR1 a  AL2 a is positive).
Therefore Ambd pro posed < Ambdchakraborty .
(II.b) For h = 0;
From the denition proposed in Section 3, the a cut of the
proposed distance measure (from Eqs. (2) and (3)) can be written
as follows:
dLa 0;

0
jAL2

sa AR2 a  AL1 a da

a  AR1 aj < AR2 a  AL1 a

0
jAL2

(11)

sa AR1 a  AL2 a da

a  AR1 aj > AR2 a  AL1 a

(12)

Now, Eq. (8) is compared with the Eqs. (11) and (12)
respectively as follows:
(i) From Eqs. (8) and (11), we can write
Z 1
Z 1
saAR2 a  AL1 a da <
saAR2 a  AL1 a AR1 a
0

 AL2 a da
(as L(a)  0, therefore AR1 a  AL2 a is positive).
Therefore Ambd pro posed < Ambdchakraborty .
(ii) From Eqs. (8) and (12), we can write
Z 1
Z 1
saAR1 a  AL2 a da <
saAR1 a  AL2 a AR2 a
0

 AL1 a da

Therefore Ambd pro posed < Ambdchakraborty .


Finally, considering the both situations, as mentioned above, we
can say that
Ambd pro posed  Ambdchakraborty :
In the paper of Chakraborty and Chakraborty [1], it was already
proved that

a da
a  AR2 aj

(since dRa AR2 a  AL1 a is positive).

dRa maxfAR1 a  AL2 a _ jAL1 a  AR2 ajg

and

95

Ambdchakraborty  Ambdvoxman :
Now, we can
 Ambdvoxman .

say

that

Ambd pro posed  Ambdchakraborty

4. Similarity measure between two generalized fuzzy numbers


4.1. Fuzzy similarity measure
It is found that the existing similarity measures have given crisp
value. A natural question at this stage is whether it is reasonable to
dene a crisp similarity measure between fuzzy objects. In real life
decision making situation, experts mostly express their opinions
linguistically. Such linguistic expressions are then captured by
fuzzy set theory and denoted as fuzzy numbers for analytic
purpose. Now, in any real eld, when two experts give their
responses in linguistic expression, the similarity between their
responses must also be expressed linguistically. Often in real world
decision making situation, it is said that the experts A and B have
much similarity in their opinion regarding a specic alternative. So
in such decision making situation, to capture this linguistic
expression much similarity, a new distance based fuzzy
similarity measure has been proposed here.
Now, in the case of fuzzy similarity measure, the similarity
measure between two experts opinion i.e., between two fuzzy
numbers is obtained as a fuzzy number. But the condence level of
each expert in his own opinion is intrinsically connected to the
linguistic expression he uses. Mathematically, when fuzzy

96

D. Guha, D. Chakraborty / Applied Soft Computing 10 (2010) 9099

similarity measure is developed then the condence level of the


experts is inherently involved. In this regard, a fuzzy similarity
measure is proposed here, which not only measures the similarity
between two fuzzy numbers, but also says how condently
experts say that the two fuzzy numbers are similar which is very

important in any real eld decision making situation (refer


Section 5 for detailed explanation). Since here, the decision
making model is constructed taking into account the condence
level of the experts, therefore the use of the fuzzy similarity
measure is reasonable.

Fig. 1. Sets of fuzzy numbers.

D. Guha, D. Chakraborty / Applied Soft Computing 10 (2010) 9099

The most obvious way of calculating similarity of fuzzy sets is


based on their distance. There are more approaches on how the
relation between the two notions in form of a function can be
expressed. According to Fan and Xie [14], the relation which has
been used here is
SA 1 ; A 2 1  dA 1 ; A 2 ;

A;
A 2  1
for 0  d

(13)

Let us consider two generalized fuzzy numbers denoted as A 1


a1 ; a2 ; b1 ; g 1 ; w1 and A 2 a3 ; a4 ; b2 ; g 2 ; w2 . Therefore, by Eq. (4)
the fuzzy distance between A 1 and A 2 is dened by
A 1 ; A 2 dL ; dR ; u; s
d
aw aw

(14)

Z w

u dLaw  max
dLa da; 0 ; and s
0
Z w



R

da da  dRaw  where w minw1 ; w2
0

After normalization, (14) will be reduced to the following form:




dLaw
dRaw
u
s
A 1 ; A 2
;
;
;
(15)
d
dRaw s dRaw s dRaw s dRaw s
Then with the help of (13), from (15), the similarity measure
between two generalized trapezoidal fuzzy numbers A 1 and A 2 will
be
A 1 ; A 2
S


1


dRaw
dL
s
u
; 1  R aw ; R
; R
;
daw s daw s daw s
aw s

dR

w minw1 ; w2

(16)

97

(1) From Fig. 1 it is clear that set 2 and set 3 are different sets of
generalized fuzzy number, but the method proposed by Lee
[17] and Chen [4] give the same similarity (from Table 1).
(2) In set 4 two different fuzzy numbers are given, but the methods
of Lee [17] and Chen [4], give the value of the similarity
measure as 1.
(3) From Fig. 1 and Table 1, we have seen that Lees [17] method
cannot calculate the similarity measure of the fuzzy numbers
given in set 5 and for the fuzzy numbers given in set 6, Lees[17]
method calculates the degree of similarity as 0.
(4) From Fig. 1 it is clear that set 7 and set 8 are different sets of
generalized fuzzy number, but Chen [4] gives the same
similarity (from Table 1).
(5) In set 9 and set 10, different sets of fuzzy numbers are given,
but by the method of Chen [4] and Chen and Chen [7] same
similarity measure is obtained.
(6) From Fig. 1, we can see that set 11 and set 12 are different sets
of generalized fuzzy numbers. However, from Table 1, we can
see that if we apply the methods proposed by Chen [4] and
Chen and Chen [7], set 11 and set 12 get the same degree of
similarity.
Now from Fig. 1 and Table 1 it is clear that in all the above cases
the proposed similarity measure calculates the similarity between
two fuzzy numbers and overcome the drawbacks of the existing
methods.
5. Fuzzy decision making technique based on the proposed
similarity measure of generalized fuzzy number (when
condence level of the expert is considered)

4.2. Properties
The above distance based fuzzy similarity measure given in
Eq. (16) satises the following properties:
(i)
(ii)
(iii)

A 1 ; A 2 2 0; 1
S
A 2 ; A 1
A 1 ; A 2 S
S
A 1 ; A 3  minfS
A 1 ; A 2 ; S
A 2 ; A 3 g
If A 1  A 2  A 3 ; then S

4.3. Comparing with existing methods


In this section, 12 sets of generalized fuzzy numbers are given
(Fig. 1) and they are used to compare the distance based fuzzy
similarity measure with the three methods presented by Chen [4],
Lee [17] and Chen and Chen [7]. A comparison between the results
of the proposed similarity measure and the results of the existing
methods is shown in Table 1. From Fig. 1 and Table 1, we can see
some drawbacks of the existing methods and some advantages of
the proposed method.

Fuzzy set theory has been used in decision making when


fuzziness present in human judgment. In a brief review of the
literature, it is seen that no methodology has been developed for
group decision making by considering the condence level of an
expert. The condence of an expert in his own response in any
decision making situation is very important. In any real eld
decision making situation when an expert gives his response to a
particular alternative, his condence level with respect to such
queries is very much important. Suppose, for example two experts
give their responses to a particular alternative: rst expert says I
have full condence in his personality being suitable for the job
under consideration. Another says that I am somewhat
condent about his personality being suitable for the job under
consideration. It is to be noted here that both the experts opine
that the candidates personality is suitable for the job but their
opinions vary in their individual condence levels. So there may be
such situations where evaluation of an alternative may be
dependent on the condence level of an expert. This is the reason

Table 1
Here a comparison of the proposed similarity measure with the existing similarity measure through the above sets of 12 examples are given.

Set
Set
Set
Set
Set
Set
Set
Set
Set
Set
Set
Set

1
2
3
4
5
6
7
8
9
10
11
12

Lees method [17]

Chens method [4]

Chen and Chens method [7]

The proposed method

0.9167
0.5
0.5
1
#
0
0.5
0.6667
0.4167
0.75
0.5
0.375

0.975
0.7
0.7
1
1
0.9
0.9
0.9
0.65
0.65
0.5
0.5

0.8357
0.42
0.49
0.8
1
0.9
0.54
0.81
0.106
0.106
0.25
0.25

(0.95, 1.0; 0.125, 0.0; 1.0)


(.65, .75; 0.125, 0.125; 1.0)
(0.6, 0.8; 0.1, 0.1; 1.0)
(0.88, 1.0; .048, 0.0; 0.8)
(1.0, 1.0; 0.0, 0.0; 1.0)
(0.9, 0.9; 0.0, 0.0; 1.0)
(0.9, 0.9; 0.05, 0.05; 1.0)
(0.9; 0.1, 0.1; 1.0)
(0.6, 0.7; 0.171, 0.24; 2/7)
(0.6, 0.69; 0.23; 0.27; 2/7)
(0.5; 0.05, 0.35; 3/5)
(0.4, 0.6; 0.32, 0.28; 2/5)

Note. # means that the similarity measure cannot calculate the degree of similarity between two generalized fuzzy numbers. The results that are not satisfactory are given in
bold and italic.

D. Guha, D. Chakraborty / Applied Soft Computing 10 (2010) 9099

98

Table 2
Linguistic variables for the ratings of alternatives.
Very poor (VP)
Poor (P)
Medium poor (MP)
Fair (F)
Medium good (MG)
Good (G)
Very good (VG)

(0; 0, 1; 1.0)
(1; 1, 2; 1.0)
(3; 2, 2; 1.0)
(5; 2, 2; 1.0)
(7; 2, 2; 1.0)
(9; 2, 1; 1.0)
(10; 1, 0; 1.0)

why condence level of an expert must include in any real life


decision making situation.
So, in this section we have presented a decision making
technique based on our proposed similarity measure. Since the
proposed similarity considers the degrees of condence of experts
opinion, therefore the use of this similarity measure is more
suitable and realistic.
In the following, we have presented the algorithm to deal with
the decision making problems considering the condence level of
the experts. The decision makers use the linguistic variables to
evaluate the candidates with respect to the attributes. Table 2 [6]
illustrates the linguistic terms and their corresponding generalized
fuzzy numbers. The various n alternatives are denoted as a1, a2, . . .,
an. For all alternatives aj the rating of the ith aspect is denoted by
fi j mi j ; ai j ; bi j ; wi j , i.e. fi j is the value of the ith criterion
function for the alternative aj; m is the number of criteria. The
weight importance of the ith attribute given by the expert is
denoted as xi. The algorithm is now presented as follows:
Step 1 Now, for each alternative aj the evaluation over all
attributes are dened as follows:
"
#
m
X
1

xi : fi j
for each j 1; 2; . . . ; n
(17)
R j Pm
i1 xi i1
and the condence level of expert regarding R j is w j
min1im wi j [7].
Now based on the similarity measure (given in Eq. (16) of
Section 4.1) a ranking process to determine the best
alternative, has been given in the steps below:
Step 2 Use the proposed similarity measure to evaluate the degree
of similarity between the over all fuzzy scores R j of the j th
alternative and the terms given in Table 2. The more value of
the similarity measure indicates more matching.
Step 3 Let us suppose, R 1 and R 2 be the aggregated fuzzy scores of
the alternative a1 and a2, respectively. Now, the ordering is
done as follows:
 First, label a1 and a2 with the terms given in Table 2 where
the similarity value (as given by 16) are maximum.
 If the labels are different, then the ordering is either
a1 > a2 or a1 < a2 according to the labeled given in Table 2.
 If a1 and a2 both are labeled with the same term, say T,
then
(i) a1 > a2 for sR 1 ; T  > sR 2 ; T 
(ii) a1 > a2 for sR 1 ; T  < sR 2 ; T 
If s R 1 ; T  s R 2 ; T  , then for ranking the alternatives
give attention to the condence level of the experts overall
opinion about an alternative and the ranking is as follows:
(i) a1 > a2 for w1 > w2 .
(ii) a1 > a2 for w1 < w2 .
Step 4 Here it is clear that the similarity measure gives a generalized
triangular fuzzy number. While comparing two fuzzy
similarity measures i.e. two generalized triangular fuzzy
numbers, it is not possible to conclude as such that one is

greater than the other. Therefore here the ranking process


[23] of generalized fuzzy numbers can be used.
Human reasoning says that similarity between two fuzzy
numbers should be fuzzy in nature as well. Therefore, an algorithm
for fuzzy decision making problem, based on the fuzzy similarity
measure, has been given in this work. Now, any approximation or
defuzzication leads to some loss of information. So, if such
methods are involved in the initial steps of the algorithm
developed then it leads to signicant loss of information. To
prevent this and to conserve the fuzziness of real world decision
making situations as far as possible, the defuzzication process has
been incorporated at a much later stage of the algorithm.
In the following, we give one example to illustrate the decision
making process based on the proposed similarity.
Example 1. Suppose, an expert gives nal decision based upon the
following criteria: Emotional steadiness (C1), oral communication
skill (C2), personality (C3). The expert uses the linguistic variables
to evaluate the candidates with respect to the attributes. The most
important thing is that, the condence level of the experts opinion
is also captured here. Now, consider our alternative set A consists
of two components a1 and a2. The expert chooses the linguistic
variables from Table 2 and presents it in Table 3, where linguistic
variables can be expressed in terms of generalized fuzzy numbers.
Now, for this problem, the condence level of the fuzzy numbers,
representing the ratings of the alternatives, is considered as 1 and
also it has been assumed here that the attributes are of equal
importance.
In the following we use the above decision making algorithm to
nd the best option among the alternatives.
Based on Eq. (17), Tables 2 and 3, the over all fuzzy numbers
combining the evaluation over all attributes, are calculated and
given below:
R 1 7; 2; 1:67; 1:0;

R 2 9:67; 1:33; 0:33; 1:0

Now, using Eq. (16), the degree of similarity between each term
of Table 2 and R 1 and R 2 is evaluated as shown in Tables 4 and 5,
respectively.
From Table 4, applying step 4, we can see that the similarity
between the over all fuzzy score R 1 and medium good has the
largest value. Therefore, the generalized triangular fuzzy number
Table 3
The ratings of candidates given by the expert under all attributes.
Criteria

Candidates

Experts response

C1

a1
a2

MG
G

C2

a1
a2

G
VG

C3

a1
a2

F
VG

Table 4
The degree of similarity between R 1 and each linguistic term given in
Table 2.
Linguistic term (say Ti)

Degree of similarity sR 1 ; T i

Very poor (VP)


Poor (P)
Medium poor (MP)
Fair (F)
Medium good (MG)
Good (G)
Very good (VG)

(0.107; 0.107, 0.191; 1.0)


(0.182; 0.182, 0.273; 1.0)
(0.3145; 0.3145, 0.3428; 1.0)
(0.17; 0.17, 0.16; 1.0)
(1.0; 1.0, 0.0; 1.0)
(0.43; 0.43, 0.523; 1.0)
(0.25; 0.25, 0.33; 1.0)

D. Guha, D. Chakraborty / Applied Soft Computing 10 (2010) 9099


Table 5
The degree of similarity between R 2 and each linguistic term given in Table 2.
Linguistic term (say Ti)

Degree of similarity s R 2 ; T i

Very poor (VP)


Poor (P)
Medium poor (MP)
Fair (F)
Medium good (MG)
Good (G)
Very good (VG)

(0.107; 0.107, 0.123; 1.0)


(0.07; 0.07, 0.18; 1.0)
(0.15; 0.0.15, 0.212; 1.0)
(0.19; 0.19, 0.3; 1.0)
(0.304; 0.304, 0.434; 1.0)
(0.63; 0.63, 0.37; 1.0)
(0.67; 0.665, 0.33; 1.0)

R 1 is translated into the linguistic term medium good. In a similar


way, the generalized triangular fuzzy number R 2 is translated into
the linguistic term very good (from Table 5). It means that, the
best option is A2.
6. Conclusion
From the normal fuzzy numbers when generalized fuzzy
numbers are extended, the condence level of a fuzzy number
varies. For generalized fuzzy numbers, the condence level of a
fuzzy number is 0  w  1, which capture more uncertainty than
normal fuzzy numbers. The proposed method, not only measures
the similarity between two generalized fuzzy numbers, but also
says how condently experts say that the two fuzzy numbers are
similar. The proposed similarity measure also provides us a useful
way to deal with fuzzy decision making problems, where experts
opinions are represented by generalized fuzzy numbers.
Acknowledgement
The rst author gratefully acknowledges the nancial support
provided by the Council of Scientic and Industrial Research, India
(Award No. 9/81(712)/08-EMR-I).
References
[1] C. Chakraborty, D. Chakraborty, A theoretical development on fuzzy distance
measure for fuzzy numbers, Mathematical and Computer Modelling 43 (2006)
254261.
[2] C. Chakraborty, D. Chakraborty, A decision scheme based on OWA operator for an
evaluation programme: an approximate reasoning approach, Applied Soft Computing 5 (2004) 4553.

99

[3] S.H. Chen, Operations on fuzzy numbers with function principal, Tamkang Journal
of Management Sciences 6 (1) (1985) 1325.
[4] S.M. Chen, New methods for subjective mental workload assessment and fuzzy
risk analysis, Cybernetics and Systems 27 (1996) 449472.
[5] S.H. Chen, Ranking generalized fuzzy numbers with graded mean integration, in:
Proceedings of the 8th International Fuzzy Systems Association World Congress,
vol. 2, 1999, pp. 899902.
[6] C.T. Chen, Extensions of the TOPSIS for group decision-making under fuzzy
environment, Fuzzy Sets and Systems 114 (2000) 19.
[7] S.J. Chen, S.M. Chen, Fuzzy risk analysis based on similarity measures of generalized fuzzy numbers, IEEE Transactions on Fuzzy Systems 11 (2003) 4556.
[8] C.H. Cheng, A new approach for ranking fuzzy numbers by distance method, Fuzzy
Sets and Systems 95 (1998) 307317.
[9] M. Delgado, A. Vila, W. Voxman, On a canonical representation of fuzzy numbers,
Fuzzy Sets and Systems 93 (1998) 125135.
[10] M. Delgado, A. Vila, W. Voxman, On a canonical representation of fuzzy numbers,
Applications Systems 94 (1998) 205216.
[11] P. Diamond, Fuzzy least squares, Information Sciences 46 (1988) 141157.
[12] D. Dubois, H. Prade, Fuzzy Sets and Systems: Theory and Applications, Academic
Press, 1980.
[13] L. Fachao, S. Lianqing, Y. Xiangdong, Q. Jiqing, The absolute value of fuzzy number
and its basic properties, The Journal of Fuzzy Mathematics 9 (2001) 4350.
[14] J. Fan, W. Xie, Some notes on similarity measure and proximity measure, Fuzzy
Sets and Systems 101 (1999) 403412.
[15] C.H. Hsieh, S.H. Chen, Similarity of generalized fuzzy numbers with graded mean
integration representation, in: Proceedings of the 8th International Fuzzy Systems Association World Congress, vol. 2, Taipai, Taiwan, Republic of China,
(1999), pp. 551555.
[16] J. Kacprzyk, Multistage Fuzzy Control, Wiley, Chichester, 1997.
[17] H.S. Lee, An optimal aggregation method for fuzzy opinions of group decision,
Proceedings of IEEE International Conference on Systems, Man, and Cybernetics 3
(1999) 314319.
[18] D.F. Li, C.T. Cheng, New similarity measures of intuitionistic fuzzy sets and application to pattern recognitions, Pattern Recognition Letters 23 (2002) 221225.
[19] Z.Z. Liang, P.F. Shi, Similarity measures on intuitionistic fuzzy numbers, Pattern
Recognition Letters 24 (2003) 26872693.
[20] W. Pedrycz, Fuzzy sets in pattern recognition: accomplishments and challenges,
Fuzzy Sets and Systems 90 (1997) 171176.
[21] L. Tran, L. Duckstein, Comparison of fuzzy numbers using a fuzzy distance
measure, Fuzzy Sets and Systems 130 (3) (2002) 331341.
[22] W. Voxman, some remarks on distances between fuzzy numbers, Fuzzy Sets and
Systems 100 (1998) 353365.
[23] Y.M. Wang, J.B. Yang, D.L. Xu, K.S. Chin, On the centroids of fuzzy numbers, Fuzzy
Sets and Systems 157 (2006) 919926.
[24] S.H. Wei, S.M. Chen, A new approach for fuzzy risk analysis based on similarity
measures of generalized fuzzy numbers, Expert Systems with Applications 36
(2009) 589598.
[25] R. Xu, C. Li, Multidimensional least-squares tting with a fuzzy model, Fuzzy Sets
and Systems 119 (2001) 215223.
[26] M.S. Yang, C.H. Ko, On cluster-wise fuzzy regression analysis, IEEE Transactions on
Systems, Man and Cybernetics Part B 27 (1997) 113.
[27] D. Yong, S. Wenkang, D. Feng, L. Qi, A new similarity measure of generalized fuzzy
numbers and its application to pattern recognition, Pattern Recognition Letters 25
(2004) 875883.

You might also like