Fuzzy Sets and Systems 157 (2006) 919 – 926

www.elsevier.com/locate/fss

On the centroids of fuzzy numbers

Ying-Ming Wanga, b,∗ , Jian-Bo Yanga , Dong-Ling Xua , Kwai-Sang Chinc
a Manchester Business School, The University of Manchester, P.O. Box 88, Manchester M60 1QD, UK
b School of Public Administration, Fuzhou University, Fuzhou, Fujian, 350002, PR China
c Department of Manufacturing Engineering and Engineering Management, City University of Hong Kong, Tat Chee, Kowloon, Hong Kong

Received 16 July 2005; received in revised form 14 November 2005; accepted 14 November 2005
Available online 7 December 2005

Abstract
In a paper by Cheng [A new approach for ranking fuzzy numbers by distance method, Fuzzy Sets and Systems 95 (1998) 307–317],
a centroid-based distance method was suggested for ranking fuzzy numbers, both normal and non-normal, where the fuzzy numbers
are compared and ranked in terms of their Euclidean distances from their centroid points to the origin. It is found that the centroid
formulae provided by the above paper are incorrect and have led to some misapplications. In this paper we present the correct centroid
formulae for fuzzy numbers and justify them from the viewpoint of analytical geometry. A numerical example demonstrates that
Cheng’s formulae can significantly alter the result of the ranking procedure.
© 2005 Elsevier B.V. All rights reserved.

Keywords: Normal fuzzy numbers; Non-normal fuzzy numbers; Centroid; Ranking

1. Introduction

In a paper by Cheng [2], a centroid-based distance method was suggested for ranking fuzzy numbers. The method
utilizes the Euclidean distances from the origin to the centroid point of each fuzzy number to compare and rank the
fuzzy numbers. Chu and Tsao [3] found that the distance method could not rank fuzzy numbers correctly if they are
negative and therefore suggested using the area between the centroid point and the origin to rank fuzzy numbers. More
recently, Abbasbandy and Asady [1] found that Chu and Tsao’s area method could sometimes lead to counterintuitive
rankings and hence suggested a sign distance method.
In real applications, we find that the centroid formulae for fuzzy numbers provided by Cheng [2] are incorrect
and have led to some misapplications such as by Chu and Tsao [3] and Pan and Yeh [7,8]. The incorrect formulae
also appeared in [1] although Abbasbandy and Asady did not use them in their approach. To avoid possible more
misapplications or spread in the future, we present in this paper the correct centroid formulae for fuzzy numbers and
justify them from the viewpoint of analytical geometry.
The rest of the paper is organized as follows. Section 2 presents the correct centroid formulae for fuzzy numbers,
derives the simplified formulae for triangular and trapezoidal fuzzy numbers, and justifies them using the knowledge
of analytical geometry. Section 3 illustrates with a numerical example the fact that the incorrect formulae by Cheng

∗ Corresponding author. Tel.: +44 161 2750788; fax: +44 161 2003505.
E-mail addresses: [email protected], [email protected] (Y.-M. Wang).

0165-0114/$ - see front matter © 2005 Elsevier B.V. All rights reserved.
doi:10.1016/j.fss.2005.11.006
920 Y.-M. Wang et al. / Fuzzy Sets and Systems 157 (2006) 919 – 926

can significantly alter the result of the ranking procedure and lead to a wrong ranking order. The paper is concluded in
Section 4.

2. The correct centroid formulae for fuzzy numbers

A fuzzy number is a convex fuzzy subset of the real line R and is completely defined by its membership function.
Let à be a fuzzy number, whose membership function fà (x) can generally be defined as [4]
⎧ L
⎪ f (x), a
x
b,
⎪
⎨ Ã

, b
x
c,
fà (x) = (1)
⎪
⎪ f R (x), c
x
d,
⎩ Ã
0 otherwise,

where 0 <

1 is a constant, f L : [a, b] → [0,
] and f R : [c, d] → [0,
] are two strictly monotonical and
à Ã
continuous mappings from R to the closed interval [0,
]. If
= 1, then à is a normal fuzzy number; otherwise, it is
said to be a non-normal fuzzy number. If the membership function fà (x) is piecewise linear, then à is referred to as a
trapezoidal fuzzy number and is usually denoted by à = (a, b, c, d;
) or à = (a, b, c, d) if
= 1, which is plotted
in Fig. 1. In particular, when b ≡ c, the trapezoidal fuzzy number is reduced to a triangular fuzzy number denoted
by à = (a, b, d;
) or à = (a, b, d) if
= 1. So, triangular fuzzy numbers are special cases of trapezoidal fuzzy
numbers.
Since f L (x) and f R (x) are both strictly monotonical and continuous functions, their inverse functions exist and
à Ã
should also be continuous and strictly monotonical. Let g L : [0,
] → [a, b] and g R : [0,
] → [c, b] be the inverse
à Ã
functions of f L (x) and f R (x), respectively. Then g L (y) and g R (y) should be integrable on the closed interval [0,
].
à  w à w à Ã
In other words, both 0 g L (y) dy and 0 g R (y) dy should exist. In the case of trapezoidal fuzzy number (see Fig. 1),
à Ã
the inverse functions g L (y) and g R (y) can be analytically expressed as
à Ã
L
gà (y) = a + (b − a)y/
, 0
y

. (2)
R
gà (y) = d − (d − c)y/
, 0
y

. (3)

which are shown in Fig. 2.

In order to determine the centroid point (x̄0 , ȳ0 ) of a fuzzy number Ã, Cheng [2] provided the following centroid
formulae:
b c d
a (xfà (x)) dx + b (x) dx + c (xfà (x)) dx
L R
x̄0 (Ã) =  b  c  d
, (4)
a (f (x)) dx + b dx + c (f (x)) dx
L R
à Ã
1 1

[ 0 (ygà (y)) dy + 0 (ygà (y)) dy]
L R
ȳ0 (Ã) = 1 L 1 R . (5)
0 (g (y)) dy + 0 (g (y)) dy
à Ã

Normal fuzzy numbers can be seen as special cases of non-normal fuzzy numbers with
= 1.
In Chu and Tsao [3], the centroid formulae were given as
b c d
a (xfà (x)) dx + b (x) dx + c (xfà (x)) dx
L R
x̄0 (Ã) =  b c d R , (6)
a (f (x)) dx + b dx + c (f (x)) dx
L
à Ã



0 (yg L (y)) dy + 0 (yg R (y)) dy
ȳ0 (Ã) = 
à 
Ã
. (7)
L
0 (gà (y)) dy + R
0 (g (y)) dy
Ã
 Y.-M. Wang et al. / Fuzzy Sets and Systems 157 (2006) 919 – 926 921

f A~L f A~R

0 a b c d x

Fig. 1. Trapezoidal fuzzy number.

gA~R

gA~L
a

0
y

Fig. 2. Inverse function of trapezoidal fuzzy number.

The above formulae were also adopted by Pan and Yeh [7,8]. However, it is found that the above formulae (4)–(7)
are
+∞  +∞the fact that formulae (4) and (6) are consistent with Yager’s ranking index [9,10] F (Ã) =
incorrect despite
−∞ g(x)f (x)/ −∞ fà (x) with the weight function g(x) ≡ x and
= 1 and the same as Murakami et al.’s ranking
à  +∞  +∞
index for normal fuzzy sets [6] z0 (Ã) = −∞ xfà (x)/ −∞ fà (x). The correct centroid formulae should be as follows:
 +∞ b c d R
−∞ xfà (x) dx a xf L (x) dx + b (x
) dx + c xfà (x) dx
x̄0 (Ã) =  +∞ = Ã
b L c d , (8)
−∞ fà (x) dx a à (x) dx
f + b (
) dx + c f R (x) dx
Ã


0 y(g R (y) − g L (y)) dy
ȳ0 (Ã) = 
à Ã
, (9)
0 (g (y) − g (y)) dy
R L
à Ã


where the denominator 0 (g R (y) − g L (y)) dy in (9) represents the area of the trapezoid in Fig. 2, while the numerator

à Ã
0 y(gà (y) − gà (y)) dy is the weighed average of the area.
R L

The main errors with formulae (4)–(7) are that
is missing from formulae (4) and (6), which makes them wrong
when
= 1, and that both formulae (5) and (7) take a positive sign for the second item in both numerator and
denominator, which is a fundamental error and makes the formulae wrong for any
values.
In particular, the natural properties of centroids are violated by formulae (4)–(7). Without loss of generality, we
consider two general fuzzy numbers: à with the membership function fà (x) defined by Eq. (1) and B̃ with its
922 Y.-M. Wang et al. / Fuzzy Sets and Systems 157 (2006) 919 – 926

membership function fB̃ (x) defined as

⎧ L
⎪ f (x − ), a + 
x
b + ,
⎪
⎨ Ã

, b + 
x
c + ,
fB̃ (x) = R (x − ), c + 
x
d + , (10)
⎪
⎪ f
⎩ Ã
0 otherwise,

where  is a non-zero constant. It is evident that B̃ is the right or left translation of à along horizontal axis. Such a
translation should make the centroid point of B̃ do exactly the same movement along the horizontal axis, but should
not change the coordinate of its centroid on vertical axis. That is to say, there should exist x̄0 (B̃) = x̄0 (Ã) +  and
ȳ0 (Ã) ≡ ȳ0 (B̃). This is one of the properties that correct centroid formulae have to possess. In addition, the changes
of
should only affect ȳ0 (Ã) and ȳ0 (B̃), and should not result in any changes of x̄0 (Ã) and x̄0 (B̃). This is another
property that correct centroid formulae should have.
Let g L (y) and g R (y) be the inverse functions of f L (x) and f R (x) and g L (y) and g R (y) the inverse functions of
à à à à B̃ B̃
f L (x) and f R (x), respectively. Since the translation does not change the shape of fuzzy number B̃, g L (y) and g R (y)
B̃ B̃ B̃ B̃
can therefore be expressed as g L (y) = g L (y) +  and g R (y) = g R (y) + . By formula (5), it can be derived that
 1 B̃ Ã
1 B̃ Ã

[ 0 (yg L (y)) dy + 0 (yg R (y)) dy + ]
ȳ0 (B̃){Cheng} = 1 L
Ã
1 R
Ã
(g
0 Ã (y)) dy + (g
0 Ã (y)) dy + 2
1  1

[ 0 (yg L (y)) dy + 0 (yg R (y)) dy]
= 1 L
Ã
1 R
Ã
= ȳ0 (Ã){Cheng}.
0 (g (y)) dy + 0 (g (y)) dy
à Ã

It is very clear that the value of ȳ0 (B̃){Cheng} varies with the value of  and ȳ0 (Ã){Cheng}  = ȳ0 (B̃){Cheng}. Such a
result is counterintuitive and unacceptable. So, formula (5) cannot be correct. As such, it can be derived from formula
(7) that



0 (ygà (y)) dy + 0 (ygà (y)) dy + 

L R 2
ȳ0 (B̃){Chu and Tsao} = 
L 
R
0 (g (y)) dy + 0 (gà (y)) dy + 2


ÃL 

0 (ygà (y)) dy + 0 (ygà (y)) dy
R
= 
L 
R = ȳ0 (Ã){Chu and Tsao},
0 (g (y)) dy + 0 (g (y)) dy
à Ã
which leads to the conclusion that formula (7) cannot be correct either.
However, by formula (9), we have the following result:



0 y(gB̃ (y) − gB̃ (y)) dy 0 y(gà (y) − gà (y)) dy
R L R L
ȳ0 (B̃) = 
R = 
R ≡ ȳ0 (Ã).
0 (g (y) − g (y)) dy 0 (g (y) − g (y)) dy
L L
B̃ B̃ Ã B̃

It is obvious that formula (9) does not cause any changes of its centroid on the vertical axis when a fuzzy number moves
along the horizontal axis.
To test whether or not the changes of
have any effect on the centriod of a fuzzy number on horizontal axis when
it changes in vertical axis, we consider a general fuzzy number: C̃ with its membership function fC̃ (x) defined as
fC̃ (x) =
fà (x), where 0 <
< 1 and à is assumed to be a normal fuzzy number. Theoretically, the fuzzy number
C̃ should have the same centroid coordinate on x-axis as the fuzzy number Ã.
By formulae (4) or (6), we have the result below:
b c d
a (xfà (x)) dx + b (x/
) dx + c (xfà (x)) dx
L R
x̄0 (C̃){Cheng} =  b  c  d
a (fà (x)) dx + b (1/
) dx + c (fà (x)) dx
L R
b c d
a (xfà (x)) dx + b (x) dx + c (xfà (x)) dx
L R
= b L c d R = x̄0 (Ã){Cheng}.
a (f (x)) dx + b dx + c (f (x)) dx
à Ã
 Y.-M. Wang et al. / Fuzzy Sets and Systems 157 (2006) 919 – 926 923

P((b+c) / 2,
) E(c+d - a,
)

F Q
0 a+b-c a b (a+d)/2 c d c+d-a x

Fig. 3. Computation of centroid of a trapezoid.

So, formulae (4) and (6) cannot be correct. However, by formula (8), it is derived that
b L c d R
a xfC̃ (x) dx + b (x
) dx + c xfC̃ (x) dx
x̄0 (C̃) =  b c d R
a fC̃ (x) dx + b (
) dx + c fC̃ (x) dx
L
b L c d R
a xfà (x) dx + b (x) dx + c xfà (x) dx
= b   ≡ x̄0 (Ã).
f L (x) dx + c dx + d f R (x) dx
a à b c Ã
It is found that formula (8) does not change the centroid coordinate of a fuzzy number on horizontal axis when it
changes its membership degrees in vertical axis. This is consistent with our intuitive judgment.
To further justify formulae (8) and (9), we present in the following another derivation from the point of view of
analytical geometry. Consider a general trapezoidal fuzzy number à = [a, b, c, d;
], whose membership function is
defined as
⎧
⎪
⎪
(x − a)/(b − a), a
x
b,
⎨

, b
x
c and 0


1,
fà (x) = (11)
⎪
⎪
(d − x)/(d − c), c
x
d,
⎩
0 otherwise.

The corresponding inverse functions g L (y) and g R (y) are determined by Eqs. (2) and (3). For this trapezoidal fuzzy
à Ã
number, the following results have been derived from formulae (4) and (5):

(d 2 − 2c2 + 2b2 − a 2 + dc − ab) + 3(c2 − b2 )

x̄0 (Ã) = , (12)
3
(d − c + b − a) + 6(c − b)
 
1 (b + c) − (a + d)(1 −
)
ȳ0 (Ã) =
1 + . (13)
3 (b + c − a − d) + 2(a + d)

By formula (7), we obtain
 
1 b+c
ȳ0 (Ã) =
1 + . (14)
3 a+b+c+d
As such, by formulae (8) and (9), we derive the results below:
 
1 dc − ab
x̄0 (Ã) = a+b+c+d − , (15)
3 (d + c) − (a + b)
 
1 c−b
ȳ0 (Ã) =
1 + . (16)
3 (d + c) − (a + b)

To determine the centroid of the trapezoidal fuzzy number à = [a, b, c, d;
] geometrically, Fig. 3 shows the geometric
graph of Ã, which is a trapezoid, and its extension used for determining the centroid of the trapezoid.
924 Y.-M. Wang et al. / Fuzzy Sets and Systems 157 (2006) 919 – 926

The two straight lines EF and PQ intersect at point G, which exactly represents the center of gravity of the trapezoid
denoted by à = [a, b, c, d;
]. In order to determine the coordinates of the point G, we write the equations of the two
straight lines EF and PQ as follows:

(x − (a + b − c))
EF: y = , (17)
(c + d − a) − (a + b − c)

(a + d − 2x)
PQ: y = . (18)
(a + d) − (b + c)
Let

(x − (a + b − c))
(a + d − 2x)
= .
(c + d − a) − (a + b − c) (a + d) − (b + c)
It follows that
 
1 dc − ab
x̄0 (Ã) = a+b+c+d − , (19)
3 (d + c) − (a + b)

which is exactly the same as Eq. (15) derived from formula (8). Substituting x = x̄0 (Ã) into either Eq. (17) or (18)
leads to the following result:
 
1 c−b
ȳ0 (Ã) =
1 + , (20)
3 (d + c) − (a + b)
which is completely the same as Eq. (16) derived from formula (9). These facts verify the correctness and rationality
of formulae (8) and (9).
For normal trapezoidal fuzzy number Ã[a, b, c, d], formula (20) can be simplified as
 
1 c−b
ȳ0 (Ã) = 1+ . (21)
3 (d + c) − (a + b)
Since triangular fuzzy numbers are special cases of trapezoidal fuzzy numbers with b = c, for any triangular fuzzy
number with a piecewise linear membership function, its centroid can be determined by
1
x̄0 (Ã) = (a + b + d), (22)
3
1
ȳ0 (Ã) =
. (23)
3
In particular, for normal triangular fuzzy number à = (a, b, d), formula (22) becomes a constant, i.e.
1
ȳ0 (Ã) = . (24)
3
In this case, normal triangular fuzzy numbers can be compared or ranked directly in terms of their centroid coordinates
on horizontal axis.
Eqs. (19)–(24) are all simplified centriod formulae for non-normal trapezoidal fuzzy numbers, normal trapezoidal
numbers, non-normal triangular fuzzy numbers and normal triangular fuzzy numbers, respectively. They are simple
enough and easy to use without the need of complicated integration calculation.

3. A numerical example

The correct formulae (8) and (9) applied to the numerical examples in Cheng [2] yield different numerical results but
they do not alter the obtained rankings of the fuzzy numbers. Nevertheless, there are cases where the wrong formulae
give wrong ranking orders. In this section, we illustrate this fact with a numerical example.
 Y.-M. Wang et al. / Fuzzy Sets and Systems 157 (2006) 919 – 926 925

Table 1
The centroids of two triangular fuzzy numbers à and B̃ obtained by different formulae

Fuzzy Centriods by Cheng’s Centroids by Chu Centroids by formulae (22)–(24)

number formulae and Tsao’s formulae
 
x̄i ȳi (x̄i )2 + (ȳi )2 x̄i ȳi x̄i ȳi x̄i ȳi (x̄i )2 + (ȳi )2 x̄i ȳi

à 2 0.5 1.5811 2 0.5 1 2 1/3 1.5275 2/3

B̃ 2 0.5114 1.5845 2 0.5105 1.021 2 9/28 1.5236 9/14

Example 1. Consider the two triangular fuzzy numbers à = (1, 2, 3; 1) and B̃ = (0.5, 2.5, 3; 27/28), whose mem-
bership functions are respectively defined as
⎧
⎪
⎪ x − 1, 1
x < 2,
⎨
1, x = 2,
fà (x) =
⎪
⎪ 3 − x, 2 < x
3,
⎩
0 otherwise,
⎧
⎪ 27 x − 0.5
⎪
⎪ , 0.5
x < 2.5,
⎪
⎪ 28 2
⎪
⎪
⎪
⎪
⎪
⎪ 27
⎪
⎨ , x = 2.5,
fB̃ (x) = 28
⎪
⎪
⎪
⎪ 27 3 − x
⎪
⎪ 2.5 < x
3,
⎪
⎪
,
⎪
⎪ 28 0.5
⎪
⎪
⎩
0 otherwise.

Table 1 shows the results obtained by Cheng’s formulae (4) and (5), Chu and Tsao’s formulae (6) and (7) and the
correct centroid formulae (8) and (9), respectively. It is very clear that both Cheng’s formulae and Chu and Tsao’s
formulae lead to an incorrect ranking order: B̃ > Ã, which is contrary to the ranking order: Ã > B̃ obtained by using
the correct centroid formulae (22)–(24). This shows the fact that incorrect centroid formulae can lead to wrong ranking
orders. So, using the correct centroid formulae is very important when comparing and ranking fuzzy numbers.

4. Conclusions

In this paper, the centroid formulae for fuzzy numbers provided by Cheng [2] were shown to be incorrect and to have
led to some misapplications. To avoid possible further misapplications or spread in the future, we presented the correct
centroid formulae for fuzzy numbers and derived their simplified expressions for trapezoidal and triangular fuzzy
numbers. We also justified the presented formulae from the viewpoint of analytical geometry. A numerical example
demonstrated that incorrect formulae could significantly alter the result of the ranking procedures and lead to a wrong
ranking order. So, the correct centroid formulae are very crucial to the distance method [2] and the area method [3]
for ranking fuzzy numbers. The simplified centroid formulae derived in this paper provide very useful computational
support to the applications of the two ranking approaches.

Acknowledgements

This research was supported by the National Natural Science Foundation of China (NSFC) under the Grant no.
70171035 and the Hong Kong Research Grant Council under the Grant no. 9040786. The authors are very grateful
to the Editor-in-Chief Didier Dubois and two anonymous referees for their constructive comments and suggestions,
which have been very helpful in improving the paper.
926 Y.-M. Wang et al. / Fuzzy Sets and Systems 157 (2006) 919 – 926

References

[1] S. Abbasbandy, B. Asady, Ranking of fuzzy numbers by sign distance, Inform. Sci., in press.
[2] C.H. Cheng, A new approach for ranking fuzzy numbers by distance method, Fuzzy Sets and Systems 95 (1998) 307–317.
[3] T.C. Chu, C.T. Tsao, Ranking fuzzy numbers with an area between the centroid point and original point, Comput. Math. Appl. 43 (2002) 111–
117.
[4] D. Dubois, H. Prade, Operations on fuzzy numbers, Internat. J. Syst. Sci. 9 (1978) 613–626.
[6] S. Murakami, H. Maeda, S. Lmamura, Fuzzy decision analysis on the development of centralized regional energy control system, Preprints
IFAC Conf. on Fuzzy Information, Knowledge Representation and Decision Analysis, 1983, pp. 353–358.
[7] H. Pan, C.H. Yeh, Fuzzy project scheduling, Proc. 12th IEEE Internat. Conf. on Fuzzy Systems, 2003, pp. 755–760.
[8] H. Pan, C.H. Yeh, A metaheuristic approach to fuzzy project scheduling, in: V. Palade, R.J. Howlett, L.C. Jain (Eds.), Knowledge-based
Intelligent Information and Engineering Systems, Springer, Berlin, Heidelberg, 2003, pp. 1081–1087.
[9] R.R. Yager, Ranking fuzzy subsets over the unit interval, Proc. 17th IEEE Conf. on Cybernetics and Society, 1978, pp. 921–925.
[10] R.R. Yager, A procedure for ordering fuzzy subsets of the unit interval, Inform. Sci. 24 (1981) 143–161.