FUZZY

sets and systems

ELSEVIER Fuzzy Sets and Systems 95 (1998) 307 317

A new approach for ranking fuzzy numbers by distance method

Ching-Hsue Cheng*
Department of Mathematics, Chinese Military Academy, Fengshan, Kauhsiung, 830, Taiwan
Received December 1994; revised August 1996

Abstract

Many ranking methods have been proposed so far. However, there is yet no method that can always give a satisfactory
solution to every situation; some are counterintuitive, not discriminating; some use only the local information of fuzzy
values; some produce different rankings for the same situation. For overcoming the above problems, we propose a new
method for ranking fuzzy numbers by distance method. Our method is based on calculating the centroid point, where the
distance means from original point to the centroid point (Yo, Y0), and the 20 index is the same as Murakami et al.'s 2o.
However, the ~7oindex is integrated from the inverse functions of an LR-type fuzzy number. Thus, we use ranking function
R(lz~) = N/(x 2 ~- )52) (distance index) as the order quantities in a vague environment. Our method can rank more than two
fuzzy numbers simultaneously, and the fuzzy numbers need not be normal. Furthermore, we also propose the coefficient
of variation (CV index) to improve Lee and Li's method [Comput. Math. Appl. 15 (1988) 887-896]. Lee and Li rank fuzzy
numbers based on two different criteria, namely, the fuzzy mean and the fuzzy spread of the fuzzy numbers, and they
pointed out that human intuition would favor a fuzzy number with the following characteristics: higher mean value and
at the same time lower spread. However, when higher mean value and at the same time higher spread/or lower mean
value and at the same time lower spread exists, it is not easy to compare its orderings clearly. Our CV index is defined as
CV = a (standard error)/# (mean), which can overcome Lee and Li's problem efficiently. In this way, our proposed
method can also be easily calculated by the "Mathematica" package to solve problems of ranking fuzzy numbers. At last,
we present three numerical examples to illustrate our proposed method, and compare with other ranking methods.
JC', 1998 Elsevier Science B.V. All rights reserved.

Kevwords: Ranking fuzzy numbers; Centroid point; Distance index; Normal (non-normal) fuzzy numbers; Coefficient of
variation (CV)

1. Introduction by Jain [15] and D u b o i s and Prade Ell]. To re-

solve the task of c o m p a r i n g fuzzy numbers, m a n y
In practical use, ranking fuzzy numbers is very authors have p r o p o s e d fuzzy ranking methods
important. F o r example, the concept of o p t i m u m which yield a totally ordered set or ranking. These
or best choice to come true are completely based on methods range from the trivial to the complex, from
ranking or comparison. Therefore, h o w to set the including one fuzzy n u m b e r attribute to including
rank of fuzzy numbers has been one of the main m a n y fuzzy n u m b e r attributes. A review and com-
problems. The concept of fuzzy numbers is presented parison of these existing methods can be found in
[7, 19, 31]. Ranking methods are classified into four
* E-mail: [email protected]. major classes according to Chen and H w a n g [7],

0165-0114/98/$19.00 ~;) 1998 Elsevier Science B.V. All rights reserved

PII S01 65-01 1 4 ( 9 6 ) 0 0 2 7 2 - 2
308 C-H. Cheng / Fuzzy Sets and Systems 95 H998) 307-317

which are listed as follows: dex is defined as a(standard error)/#(mean), which

(1) Preference relation can solve Lee and Li's problem efficiently. In this
(a) Degree of optimality (as in [3, 27, 2]). way, our proposed method also can be easily cal-
(b) Hamming distance (as in [28, 16, 24, 17]). culated by the "Mathematica" package developed
(c) e-cut (as in [1, 4, 21]). to solve problems of ranking fuzzy numbers• At
(d) Comparison function (as in [12, 26, 10]). last, we present three numerical examples to illus-
(2) Fuzzy mean and spread trate our proposed method, and compare with
Probability distribution (as in [12, 26, 10]). other ranking methods•
(3) Fuzzy scoring The purpose of this paper is to present a com-
(a) Proportion to optimal (as in [19]). parison method for fuzzy numbers based on dis-
(b) Left/right scores (as in [,14, 15, 5, 6]). tance index and CV index• Section 2 presents basic
(c) Centroid index (as in [23, 29]). concepts and definitions of fuzzy mathematics. Sec-
(d) Area measurement (as in [20, 30]). tion 3 introduces how to calculate centroid point
(4) Linguistic expression and use our distance index to rank fuzzy numbers•
(a) Intuition (as in [13]). The procedure for using CV index to improve Lee
(b) Linguistic approximation (as in [25]). and Li's method is presented in Section 4. Three
Some ranking methods assume that the member- simple numerical examples are shown in Section
ship function is normal, but in many cases, limita- 5 to illustrate our method of distance index.
tion to the normal membership function is not Finally, concluding remarks are given in Section 6.
adequate. In [20], Liou and Wang show that
Chen's method [5] can give non-logical result.
However, some are counterintuitive, not discrimi- 2. Preliminaries
nating or complex; and most of them consider only
one point of view on comparing fuzzy quantities. A fuzzy number is a fuzzy subset in support
Further, many of them produce different rankings R (real number) which is both "normal" and "con-
for the same problem. About centroid index, vex", where supp(A) = {x e R I#x > 0} [8].
Yager's Yo index may be seen as a general formula Normality implies that: 3x e R, Uxp,t(x) = 1,
for calculating both Xo and jTo given different that is, the maximum value of the fuzzy set ,4 in R
functions 9(x). However, only when 9(x)=x, is 1. Therefore, the non-normal fuzzy number is:
Murakami et al.'s fro can be calculated, and when Vx ~ R, Maxx{/~(x)} < 1.
O(x) = #x(x)/2, Murakami et al.'s Yo can be de- For convenience, the fuzzy number can be de-
rived. Thus, Murakami et al.'s method is not logi- noted by [a, b, c, d; 1], and the membership func-
cally sound either. tion f~ of the fuzzy number .4 = [a, b, c, d; 1] can
For overcoming the above-mentioned problems, be expressed as
we propose ranking of fuzzy numbers by distance
L
method based on calculating the centroid point, f~ (x), a ~ x ~ b ,
where distance means from original point to the 1, b~x~c,
centroid point. Thus, we use the ranking function f~= a (1)
R(A) = x ~ + 372 as order quantities in a vague f} (x), c ~ x ~ d ,
environment, which can rank more than two fuzzy 0, otherwise,
numbers simultaneously, and need not be normal.
Furthermore, we also propose a coefficient of vari- where f~" [a, b] --* [0, 1] and f~R : [-C , d] ~ [-0, 1].
ation (CV) to improve Lee and Li's method [19]. Since f ~ : [a, b] ~ [0, 1] is continuous and strictly
Lee and Li's method ranks fuzzy numbers based on increasing, the inverse function o f f ) exists• Sim-
two criteria, namely, the fuzzy mean and the fuzzy ilarly, f f : [c, d] -~ [0, 1] is continuous and strictly
spread of the fuzzy numbers. However, when the decreasing, the reverse function o f f j also exists.
• • - R •

fuzzy mean and fuzzy spread are different, it is not The inverse functions o f f ) and f~R are denoted
R
easy to compare its orderings clearly. Our CV in- by g} and 9~, respectively• Sincef~: [-a, b] --* [0, 1]
 C.-H. Cheng / Fuzzy Sets and Systems 95 (1998) 307 317 309

3 5 7

Fig. 1. The ordering of fuzzy numbers is/T </~ < C </).

is continuous and strictly increasing, g}" [0, 1] each fuzzy number. However, in Murakami et al.'s
[a,b] is also continuous and strictly increasing. example, all 35/values are same. Thus, the ff value
Similarly, if f ~ : [c, d] ~ [0, 11 is continuous and seems to be the only rational index for comparing
strictly decreasing, then gff" [0, 11--* [c, d] is con- fuzzy numbers.
L R
tinuous and strictly decreasing; g~i and g.i are con- From Fig. 1, an ~ value on the horizontal axis is
tinuous on a closed interval [0, 1] and they are the most important index for ranking fuzzy num-
integrable on [0, 1]. That is, both So1 g~(y)
L dy and bers. However, a 35 value on the vertical axis is an
g~i dy exist. aid index, only in special cases the 35 value is an
important index for ranking fuzzy numbers (such as
all 2 values are equal or left and right spreads are
3. Ranking fuzzy numbers by distance method based same for all fuzzy numbers).
on calculating centroid point For satisfying the above condition, and over-
coming Murakami et al.'s example, we propose
Method of ranking with centroid index finds the ranking fuzzy numbers by distance method based
geometric center of a fuzzy number, A. Each geo- on calculating both ~ and 35 values. Our method
metric center corresponds to an ~ value on the can be introduced step by step as follows:
horizontal axis and an ~f value on the vertical axis. (A) From Eq. (1), using f ~ : [a, b] ~ [0, 1] and
The ordering of the fuzzy number is conducted f ~ : [c, d] --, [0, 1] to derive the inverse functions of
L
either on ff alone or with the help of both ff and f ~ and f ~ , which are g2:[O,1]~[a,b] and
R
35 values. Yager's method calculates for each fuzzy g~i'[0, 1] ~[c,d], respectively. For easy calcu-
number only Y value, and Murakami, Maeda, and lation, we can use "Mathematica" package to ac-
Imamura's method calculates both ff and .9 for complish this. For a trapezoidal fuzzy number
310 C.-H. Cheng / Fuzzy Sets and Systems 95 (1998) 307-317

/~ = [a,b,c,d; 1], its membership function f~ is By Eq. (3),

given by
Xo(/T)
x--a
a~x~b,
b-a' wff[xxb-~aJdX+wf;xdx+w£alxXc-ddJdX
1, b~x~c,
f~(x) = x -- d (2) W jafbX--adX
b _ a + w fb dx + w f f X c ~ d dx
c~x~d,
c-d'
c d R
O, otherwise. Sb,(xfL) dx + S b x d x + Sc(xfA)dX
a R ' (6)
Ib. (f~)dx + ~; dx + ~c (fA)dx
Clearly, the inverse functions of f ~ and fff are
L R the 2o value is the same as for normal fuzzy
gy, = a + (b - a)y and g2 = d + (c - d)y.
(B) The centroid point (So, JTo) for a fuzzy num- numbers.
ber/T is defined as But,

2 o ( ~ ) = [ . ~ ( x f L ) d x + [.;xdx + [.~(xfff)dx ~ g~(wy) dy

b L c
~, (f~) dx + Sbdx + S ~ ( f f ) d x
(3)
= [a + (b - a)(wy)/w] dy = O~(Y) dy,
yo(A) S°~(yg~-) dy + So1 (Yg.~)
R dy
= So (g2) dy + S o
1 (gx)
. dy ,
] g~(wy) dy
where index ~o is the same as Murakami et a1.'s So.
For Eq. (3), we can use the "Mathematica" package
to obtain the value of the centroid point (2o, 370).
(C) Calculation of ranking function: We pro-
=
;2 [d + (c - d)(wy)/w] dy =
i2" 0;~ (Y)dy

pose the distance index between the centroid point and

(So, 29o) and original point, i.e.,
(wy) g~(wy) dy = w [dy + (b - a)(wyZ)/w] dy
R(/~) = ,,,/(2o) 2 + (370)2 . (4)

(D)~Ranking fuzzy numbers: For any fuzzy num- = w f 1 (y)g2(y)dy,

L
bers Ai, Aj E S, where S = {/~i,/~2, ... , A,} is a set 0
of convex fuzzy numbers. The ranking fuzzy num-
ber has thefollowing properties: (wy)g~ (wy)dy = w [dy + (c - d)(wy2)/w] dy
(1) If R(AI) < R(A;), then J-i < / ~ j ,
(2) if R(A,) = R(/~), then/~i = Ai, and = w (y)g~(y) dy.
(3) if R(A~) > R(J.j), then A~ >/11.
Finally, for non-numbers, it is given by Therefore,
1 L
w(x - a) w [~o (Y)g~,(Y) dy + [.~o(y)gR(y) dy]
b-a ' a<~x<~b, -
37o(.4) - 1 L
~oO~(y)dy 1 R
+ Iog~(y)dy , (7)

w, b <~ x <~ c and 0 ~ < w 4 1 , where w is the adjustment factor for the 370 value.
f~(x) = wtx - d) (5)
c-d ' c<~x<~d,
0, otherwise. 4. Using the CV index to improve Lee and Li's
method
The reverse functions of fL and f R are ggL =
a + (b - a)y/w and g~ = d + (c - d)y/w, respec- In many fuzzy multiple criteria decision making
tively, where y e [0, w]. problems, the final scores of alternatives are
 C.-I-L Cheng/ Fuzzy Sets and Systems 95 (1998) 307-317 31 t

represented in terms of fuzzy numbers. In order to We have

choose a best alternative, we need a method for
building a crisp total ordering from fuzzy numbers.
Many methods for ranking of fuzzy numbers have
2p(j`)=fs,x)xf~,~(x)dX/fs,~(ux(x))fdx, (12)

been suggested. Each method appears to have some

advantages as well as disadvantages [7]. In fuzzy #p(j`) = [ (fs(~) x2(~i(x))2 dx /'/
multiple criteria decision making problem, many
triangular fuzzy numbers can intuitively rank its
ordering by drawing its curves. If its ordering can-
fs(~i)(,a~.(x))2dx)-(fp(j`))2 ] 1/2 . (13)
not be ranked by figures, we can use many other
When J` is a triangular fuzzy number, Eqs. (12)
methods of ranking fuzzy numbers. In this section,
and (13) may be rewritten as
firstly, we summarize Lee and Li's method. Then,
we propose the CV index to improve Lee and Li's 1 (1 + 2m + n), (14)
method.
ap(j`) =~0 (3/2 + 4m2 + 3n2 -41m -21n -4mn).
(15)
4.1. Lee and Li's method
4.2. The index of coefficient of variation
Lee and Li proposed the use of generalized mean
and standard deviation based on the probability From the concept of statistics, the standard devi-
measures of fuzzy events to rank fuzzy numbers. ation and mean value cannot be the sole basis for
The method ranks fuzzy numbers based on two comparing two fuzzy numbers, respectively. Fur-
different criteria, namely, the fuzzy mean and the thermore, according to Lee and Li, higher mean
fuzzy spread of the fuzzy numbers. They assume value and at the same time lower spread is ranked
two kinds of probability distributions for fuzzy higher. However, when higher mean value and at
events and derive corresponding indices as follows: the same time higher spread/or lower mean value
(1) Uniform distribution: f ( j ` ) = 1/]j`] and and at the same time lower spread, it is not easy to
J`e U. compare the orderings clearly. Therefore, we pro-
Its mean flu(J,) and standard deviation #u(j`) are pose an efficient index, that is, using the coefficient
defined as of variation to improve Lee and Li's method.
The coefficient of variation is a relative measure
fu(j`) = f s ( i ) x l t i ( x ) d x / f s , ~ , " (x)dx (s) of dispersion. It relates the standard deviation and
the mean by expressing the standard deviation as
a percentage of the mean. It is defined as
cru(j`) = I (is(%) x2#%(x) dx /
CV = a(standard error)/]/~l(mean),

fsci)t~i(x)dx) - (Xu(J`))211/2, (9)

where/~ va 0 and a > 0. (16)
For comparing fuzzy numbers by our CV index
where S(J`) is the support of fuzzy number J`. method, firstly, we can calculate mean value and
When the fuzzy number J` is a triangular, Eqs. (8) standard deviation by Lee and Li's method. Fi-
and (9) may be rewritten as nally, the fuzzy number with smaller CV is ranked
higher.
~ u ( j ` ) = 1 (1 + m + n), (10) The following example was presented to illus-
GU(J`) = ~88 (/2 -t- IT/2 -1- rt 2 -- Im -- In -- mn), (1 1) trate our CV index method, which can improve Lee
and Li's method efficiently.
where l = inf S(J,),/~,(m) = 1, n = sup S(J` ).
(2) Proportional distribution: f(j`) -- klli(x), Example 4.1. Consider the triangle fuzzy num-
J` ~ P, where k is the proportional constant. bers U1, U2, which can be expressed as follows
312 C-I-£ Cheng / Fuzzy Sets and Systems 95 (1998) 307 317

U2

115 1 714 2

Fig. 2. Two fuzzy numbers U1 and t]z.

(indicated in Fig. 2): 5. Three numerical examples for illustrating

our distance index
{2 0~<1~<1,
fv'= '-x, l~<x~<2,
Example 5.1. In Fig. 3, three triangular fuzzy num-
f¼(5x--1), 1/5~<x~< 1, bers U1, U2, U3 from [17] are shown.
fD2=[½(7_4x), 1~<x_<2..~4.

From Eqs. (12)-(16), we can calculate its mean x -0.2 0.2 <~ x < 0.3,
0.1 '
value if, standard deviation a and CV values, which
f~= 1, x=0.3,
are listed in Table 1.
From the results of Table 1, the ~ criterion is 0.5-x 0.3 ~< x ~< 0.5,
t] 2 < Ill, and the a criterion is also 02 • U1. How- 0.2 '
ever, Lee and Li's criterion is higher mean value and
at the same time lower spread. Clearly, we cannot x -0.17
0.15 ' 0.17~<x<0.32,
easily compare its orderings by Lee and Li's
method. Therefore, we can use our CV index to iv2 1, x = 0.32,
improve its shortcomings. F r o m Table 1, we can 0.58 - x
easily rank its orderings by CV values, and the CV 0.26 ' 0.32~<x~<0.58,
value of fuzzy number U2 is smaller than the fuzzy
number UI. Therefore, its ordering is t] 2 > t] 1. x -0.25
0.15 ' 0.25~<x<0.4,
Actually, from [9], many triangular fuzzy num-
bers can be intuitively ranked according to their f~ = 1, x = 0.4,
ordering by drawing their curves. Clearly, the 0.7 - x
ordering is U2 > U1 in this example. 0.3 ' 0.4 ~ x ~< 0.7.
 C.-H. Cheng / Fuzzy Sets and Systems 95 (1998) 307-317 313

0 0.1 0.3 0.5 0.? 0.9 1

Fig. 3. Three triangular fuzzy numbers 01, 02, 03-

Table 1 Table 3
The ~, a and CV values for fuzzy numbers 01, U2 The centroid point (:el, 371) and R(UI) =

Uniform distribution Proportional distribution

xi Yi R(Oi) = x/(2i) 2 + (yi)2
2 a CV 2 cr CV
01 0.333 0.4872 0.590
01 1 0.1667 0.1667 1 0.1 0.1 02 0.357 0.4868 0.604
U2 0.9833 0.1001 0.1018 0.9875 0.0563 0.0571 U3 0.450 0.4857 0.662

Table 2 Example 5.2. In Fig. 4, the triangular fuzzy number

The inverse function o f f ~ and f ~ A1 = (3, 5, 7; 1), non-normal triangular fuzzy num-
L gn ber ¢zl2 = (3, 5, 7; 0.8), a n d trapezoidal number /~1
= (5,7,9, 10; 1), non-normal trapezoidal fuzzy
Ua 0.2 + 0.1y 0.5 - 0.2y numbers/32 = (6, 7, 9, 10; 0.6) and/J3 = (7, 8, 9, 10; 0.4)
~'2 0.17 + 0.15y 0.58 -- 0.26y are from [20].
U3 0.25 + 0.15y 0.7 - 0 . 3 y
x-3
2 ' 3~<x<5,
1, x=5,
From step (A), we can derive the inverse function
7--x
o f f ~ and f ~ in Table 2. 2 ' 5<x~<7,

f~2={
By Eqs. (3) and (4), we can obtain the last results
by "Mathematica" package in Table 3.
x-3
From Table 3, the ordering of fuzzy numbers is 0.8-- 3~<x<5,
~-~1 < [-~2 < [-~3" This is an example to show that in 2 '
Murakami et al.'s method, all .90, values are 0.8, x =5,
same. However, in our method, all .9~, values are 7--x
0.8-- 5<x~<7,
different. 3 '
314 C.-H. Cheng / Fuzzy Sets and Systems 95 (1998) 307-317

B l

0,8 L

0.~ /

0,4

0.2

I
0 1 3 5 ? 9 10

Fig. 4. Five fuzzy numbers A1, ~z12and B,, B2, B 3.

x-5 Table 4

f
2 ' 5~<x<7, The inverse functions o f f ~ , f ~ and f ~ , f ~

1, 7~<x~<9, g~L and g~L g~R and g~i

R

10 - x
1 , 9<x~<10, A1 3 + 2y 7 -- 2y
A2 3 + 2/0.8y 7 -- 2/0.8y
/~1 5 +2y 10 - y
0.6 x - 6 B2 6 + 1/0.6y 10 - 1/0.6y
1 ' 6~<x<7,
/~3 7 + 1/0.4y 10 - 1/0.4y
f~2 = 0.6, 7~<x~<9,
10-x
0.6--, 9<x~<10,
1
Table 5
The centroid point and values (xl,)7i) and R 0 ( ) =
x-7 +/;,) 2
0.4-- 7~<x<8,
1 '

fB3 ~- 0.4, 8~<x~<9, xi )~i R(X) = x/(£1) 2 + 071)2

0.5 - x
0.4-- 9<x~<10. A1 5 0.5 5.03
0.2 ' A2 5 0.4 5.02
/~1 7.714 0.505 7.73
From step (A), we can derive the inverse functions /~2 8 0.3 8.01
off~, f ~ a n d f ~ , f~R in Table 4. By Eqs. (3)and (4), /~3 8.5 0.2 8.50
we can obtain the last results by "Mathematica"
package in Table 4. From Table 5, the ordering of
fuzzy numbers is/T2 </T1 < B1 < /~2 < B3. This is
an example to show that in Liou and Wang's Example 5.3. In Fig. 5, with triangular fuzzy num-
method, ,i2 = A1 for all e, e e [0, 1]. However, our bers -41 = (0.4, 0.5, 1; 1),/~2 = (0.4, 0.7, 1; 1) and -43 =
method can rank its ordering. (0.4,0.9,1;1). Their membership functions are
 C.-H. Cheng / Fuzzy Sets and Systems 95 (1998) 307-317 315

0A 0.3 0.5 0.7 0.9 1

Fig. 5. Three triangular fuzzy numbers A1,/~2 and/la.

defined as Table 6
• he inverse function of/ and
x -0.4
- - , 0.4~<x < 0 . 5 , L R
0.1 g~ g~
fAS1(N) = 1, X =0.5, A~ 0.4 +0.1y 1 --0.5y
1--x 0.5 < X ~< 1, /i 2 0.4 + 0.3y 1 -0.3y
0.5 ' A3 0.4 +0.5y 1 -0.1y

x - 0.4
- - , 0.4 ~< x < 0.7,
0.3 Table 7
f~2(x) = 1, x =0.7, The centroid point (-fl,.Pl)and R(A) = ~ + 071)2
1-x
0.7~<x~< 1, 21 351 R(,4) = ~ + ()~,)2
0.3 '

x - 0.4 A1 0.633 0.472 0.790

- - , 0.4 ~< x < 0.9, Az 0.7 0.5 0.860
0.5 A3 0.767 0.521 0.927
fx~(x) = 1, x =0.9,
1--x
0.9 < x ~ < 1.
0.1 '
ranked higher. Due to this, all fuzzy sets have the
same spread. Therefore, we can directly rank their
Similarly, we can derive the inverse function o f f ~ , ordering by 2i.
and f ~ in Table 6, and obtain the last results by
" M a t h e m a t i c a " p a c k a g e in Table 7.
F r o m Table 4, the ordering fuzzy n u m b e r s is 6. Conclusions

According to Lee and Li [18], a fuzzy n u m b e r Yager's 2o index measures the general m e a n of
with larger m e a n and smaller spread should be fuzzy numbers. It is not surprising to see that the 2o
316 C.-H. Cheng / Fuzzy Sets and Systems 95 (1998) 307-317

index alone provides very poor discrimination abil- [-8] C.H. Cheng and D.L. Mon, Fuzzy system reliability analy-
ity. Yager's :~o index may be seen as a general sis by confidence interval, Fuzzy Sets and Systems 56
formula for calculating both So and 35o given differ- (1993) 29-35.
[-9] C.H. Cheng, Evaluating weapon systems using ranking
ent functions O(x). However, only when g(x)= x,
fuzzy numbers, submitting to Fuzzy Sets and Systems
Murakami et al.'s ~o can be calculated, and when (1996).
g(x) = ip~(x),
1 Murakami et al.'s 350 can be derived. [10] M. Delgado, J.L. Verdegay and M.A. Villa, A procedure
Thus Murakami et al.'s method is not logically for ranking fuzzy numbers using fuzzy relations, Fuzzy
sound either. Sets and Systems 26 (1988).
[,11] D. Dubois and H. Prade, Operations on fuzzy numbers,
An ~o value on the horizontal axis is the most
lnternat. J. Systems Sci. 9 (1978) 631 626.
important index for ranking fuzzy numbers. But, [12] D. Dubois and H. Prade, Ranking of fuzzy numbers in
a 350 value on the vertical axis is an aid index, only the setting of possibility theory, Inform. Sci. 30 (1983)
in special cases (such as all values are equal or left 183 224.
and right spread are same for all fuzzy numbers), 350 [13] J. Efstathiou and R. Tong, Ranking fuzzy sets using
linguistic preference relations, Proc. lOth lnternat. Symp.
is an important index for ranking fuzzy numbers.
on Multiple-Valued Logic, Northwestern University,
For satisfying the above condition, and overcoming Evanston (1980) 137 142.
Murakami et al.'s example, we have proposed [,14] R. Jain, Decision making in the presence of fuzzy variables,
a new approach for ranking fuzzy numbers, which IEEE Trans. Systems Man Cybernet. SMC-6 (1976)
are based on calculating both 20 and 350 values and 698-703.
distance index. [15] R. Jain, A procedure for multi-aspect decision making
using fuzzy sets, Internat. J. Systems Sci. 8 (1978)
Further, we have also presented the CV index to 1-7.
improve Lee and Li's method. When higher mean [16] E.E. Kerre, The use of fuzzy set theory in electrocardio-
value and at the same time higher spread/or lower logical diagnostics, in: M.M. Gupta and E. Sanchez,
mean value and at the same time lower spread is Eds., Approximate Reasonin 9 in Decision Analysis (1982)
present it is not easy to compare its orderings 277-282.
[17] W. Kolodziejczyk, Orlovsky's, concept of decision-making
clearly. Therefore, we can efficiently use CV index
with fuzzy preference relation - further results, Fuzzy Sets
to rank its ordering, and the CV criterion is ranked and Systems 19 (1986) 11-20.
higher with smaller CV value. [18] P.J.M. Laarhoven and W. Pedrycz, A fuzzy extension of
Saaty's priority theory, Fuzzy Sets and Systems 11 (1983)
229 241.
References [19] E.S. Lee and R.L. Li, Comparison of fuzzy numbers based
on the probability measure of fuzzy events, Comput. Math.
1-1] J.M. Adamo, Fuzzy decision trees, Fuzzy Sets and Systems Appl. 15 (1988) 887-896.
4 (1980) 207-220. [20] T.-S. Liou and M.-J.J. Wang, Ranking fuzzy numbers
[-2] J.F. Baldwin and N.C. Guild, Comparison of fuzzy sets on with integral value, Fuzzy Sets and Systems 50 (1992)
the same decision space, Fuzzy Sets and Systems 2 (1979) 247 255.
213-231. [21] S. Mabuchi, An approach to the comparison of fuzzy
[-3] S.M. Baas and H. Kwakernaak, Rating and ranking of subsets with an c~-cut dependent index, IEEE Trans.
multiple aspect alternative using fuzzy sets, Automatica 13 Systems Man Cybernet. SMC-18 (1988) 264 272.
(1977) 47-58. [22] C. McCahone, Fuzzy set theory applied to production and
[4] J.J. Buckley and S. Chanas, A fast method of ranking inventory control, Ph.D. thesis, Department of Industrial
alternatives using fuzzy numbers (Short communications), Engineering, Kansas State University (1987).
Fuzzy Sets and Systems 30 (1989) 337-339. [23] S. Murakami, S. Maeda and S. Imamura, Fuzzy decision
[-5] S.H. Chen, Ranking fuzzy numbers with maximizing set analysis on the development of centralized regional energy
and minimizing set, Fuzzy Sets and Systems 17 (1985) control; system, IFAC Syrup. on Fuzzy Inform. Knowledge
113 129. Representation and Decision Anal. (1983) 363 368.
[-6] S.-J. Chen and C.-L. Hwang, Fuzzy scoring of fuzzy num- [24] K. Nakamura, Preference relation on a set of fuzzy utilities
ber A direct comparison index, unpublished paper as a basis for decision making, Fuzzy Sets and Systems 20
(1989). (1986) 147-162.
[-7] S.-J. Chen and C.-L. Hwang, Fuzzy multiple attribute [-25] R.M. Tong and P.P. Bonissone, Linguistic solutions to
decision making methods and applications, Lecture Notes fuzzy decision problems, in: H.J. Zimmermann, Ed.,
in Economics and Mathematical Systems (Springer, New TIMS/Studies in the Management Science, Vol. 20
York, 1992). (Elsevier, Amsterdam, 1984) 323-334.
 C-H. Cheng / Fuzzy Sets and Systems 95 (1998) 307 317 317

[26] Y. Tsukamoto, P.N. Nikiforuk and M.M. Gupta, On the [28] R.R Yager, On choosing between fuzzy subsets, Kyber-
comparison of fuzzy sets using fuzzy copping, in: netes 9 (1980) 151-154.
H. Akashi, Ed., Control Science and Technology for [29] R.R. Yager, On a general class of fuzzy connectives, Fuzzy
Proyress of Society (Pergamon Press, New York, 1983) Sets and Systems 4 (1980) 235-242.
46-51. [30] R.R. Yager, A procedure for ordering fuzzy subsets of the
[27] S.R. Watson, J.J. Weiss and M.L. Donnell, Fuzzy decision unit interval, Inform. Sci. 24 (1981) 143 161.
analysis, IEEE Trans. Systems Man Cybernet. SMC-9 [31] H.J. Zimmermann, Fuzzy Set, Decision Makin 9 and Expert
(1979) 1 9. System (Kluwer, Boston, 1987).