Interval Type-2 Fuzzy Multi Criteria Decision Making
Based on Intuitive Multiple Centroid
Ku Muhammad Naim Ku Khalif 1, Alexander Gegov 2, Ahmad Syafadhli Abu
Bakar 3, 4, Noor Zuraidin Mohd Safar 5
1
Centre for Mathematical Sciences, Universiti Malaysia Pahang, Malaysia
2
School of Computing, University of Portsmouth, United Kingdom
3
Centre for Foundation Studies in Science, Universiti Malaya, Malaysia
4
Centre of Research for Computational Sciences and Informatics in Biology, Bioindustry,
Environment, Agriculture and Healthcare (CRYSTAL), Universiti Malaya, Malaysia
5
Faculty of Computer Science and Information Technology, Universiti Tun Hussien Onn
Malaysia, Malaysia
kunaim@ump.edu.my, alexander.gegov@port.ac.uk, ahmadsyafadhli@um.edu.my,
zuraidin@uthm.edu.my
Abstract. This paper aims to introduce fuzzy multi criteria decision making
model using consistent fuzzy preference relations and fuzzy technique for order
performance by similarity to ideal solution sets that is incorporated with
intuitive multiple centroid defuzzification in the context of interval type-2
fuzzy. The implementation of interval type-2 fuzzy sets is taken into
consideration, where it has more authority to provide more degree of freedom
in representing the uncertainty of human based decision making problems. It
also highlights the combination of interval type-2 fuzzy sets with multi criteria
decision making techniques allow the use of fuzzy linguistic by considering the
need of human intuition in decision making problems under uncertain
environment. Numerical example is included to illustrate the proposed model.
The proposed model is importantly needed to validate using sensitivity analysis
in order to analyse the quality and robustness of the model in giving the most
promising alternative with respect to resources. The results show that it is
highly practical to use the proposed model in decision making evaluation.
Keywords: Fuzzy multi criteria decision making, Consistent fuzzy preference
relations, Fuzzy TOPSIS, Interval type-2 fuzzy sets, Intuitive multiple centroid,
uncertainty
1 Introduction
Uncertainty and fuzziness are well-known phenomena in many applications areas in
science and engineering, where are often not crisp but there exist various degree of
membership grade that practical automatically occurs in decision making problems.
Type-2 fuzzy sets are appropriately tools for uncertainty or approximate reasoning
modelling. It has more authority to provide more degree of freedom in representing
the uncertainty of human based decision making problems. Klir and Yuan claim that
the type-1 fuzzy sets inly describe imprecision not uncertainty [1]. On particular
motivation for the further interest in type-2 fuzzy sets that its’ provide a better scope
for modelling uncertainty than type-1 fuzzy sets [2]. According to Karnik and
Mendel, they claim that type-2 fuzzy sets can be characterised as fuzzy membership
function where the membership value for type-2 fuzzy sets is in interval form [0,1],
unlike type-1 fuzzy sets where the membership value is a crisp value in [0,1] [3].
Defuzzification plays important role in the performance of fuzzy systems’
modelling techniques. Defuzzification process is guided by the output fuzzy subset
that one value would be selected as a single crisp value as the system output. While
much of the literature discuss there are variety of defuzzification methods have
largely developed. Though, each of them have difference performance in difference
applications and there is a general method can satisfy the performance in all
conditions which is centroid method [4]. Centroid defuzzification methods of fuzzy
numbers have been explored for the last decade that commonly used and have been
applied in various discipline areas. The computational complexity of type-2 fuzzy sets
is very difficult to handle into practical applications because of characterised by their
footprint of uncertainty [5].
In literature, most of the hybrid MCDM model combined two techniques in order
to tackle the evaluation of criteria and the evaluation of alternatives respectively. The
evaluation process of criteria and alternatives play important role in MCDM
techniques requirements. To identify the best decision to be made among the various
alternatives with several criteria, the methodology has to study the preferences among
the criteria to make sure the weights of criteria are reliable enough to be implemented
in the selection of alternatives. In this paper, the hybrid of consistent fuzzy preference
relations and fuzzy technique for order of preference by similarity to ideal solution
(TOPSIS) using new centroid defuzzification for interval type-2 fuzzy sets is
proposed in dealing with uncertainty events. The major weaknesses of classical
TOPSIS are in not providing for weight elicitation, and consistency checking for
judgments’ evaluation. Hence, in this paper, the authors consider the fuzzy TOPSIS’s
employment that has been significantly restrained by the human capacity for
information processing. Sensitivity analysis [6] is applied to validate the proposed
model. It can effectively contributes to making accurate decisions by assuming that a
set of weights for criteria or alternatives then obtained a new round of weights for
them, so that the efficiency of alternatives has become equal or their order has
changed.
The rest of this paper is organised as follows: Section 2 discusses the theoretical
preliminaries of fuzzy set theory and generalised trapezoidal fuzzy numbers. This is
then preceded to the proposed work of integrated fuzzy MCDM model that consist of
consistent fuzzy preference relations and fuzzy TOPSIS using intuitive multiple
centroid defuzzification in Section 3. Section 4 discusses the case study and results
that illustrated the proposed model and validation processes using sensitivity analysis.
Finally, Section 5 gives the conclusion.
2 Theoretical Preliminaries
In this section, the authors briefly review some definitions of interval type-2 fuzzy set
that are illustrated as follows.
A.
Interval Type-2 Fuzzy Set
Definition 1[7]: A type-2 fuzzy set A in the universe of discourse X represented by
the type-2 membership function, . If all ( x, u ) 1 , then A is called an interval
A
type-2 fuzzy sets. An interval type-2 fuzzy set can be considered as a special case of
type-2 fuzzy sets, denoted as follows.
A
1 /( x, u) , where J
x
0,1
(1)
xX uJ x
Definition 2 [7]: The upper and lower membership functions of an interval type-2
fuzzy set are type-1 fuzzy sets membership functions, respectively. A trapezoidal
interval
type-2
fuzzy
set
can
be
represented
by,
~ ~
~
~
~
~ ~
Ai ( AiU , AiL ) ((aiU1 , aiU2 , aiU3 , aiU4 ; H1 ( AiU ), H 2 ( AiL )), (aiL1 , aiL2 , aiL3 , aiL4 ; H 1 ( AiL ), H 2 ( AiL )))
~
~
where can be depicted in Fig. 1 [8]. The AiU and AiL are type-1 fuzzy sets,
aiU1 , aiU2 , aiU3 , aiU4 , aiL1 , aiL2 , aiL3 and aiL4 are the reference points of the interval type-2
~
fuzzy sets A , H j ( AiU ) denote the membership value of the element a iU( j 1) in the
~L
~
upper trapezoidal membership function AiU , 1 j 2 , H j ( Ai ) denotes the
membership value of the element a iL( j 1) in the lower trapezoidal membership
~
~
~
~
function AiL , 1 j 2 , and for H1 ( AiU ) [0,1] , H 2 ( AiU ) [0,1] , H1 ( AiL ) [0,1] ,
~
~
H 2 ( AiL ) [0,1] and 1 i n , H 2 ( AiU ) [0,1] .
A~ ( x)
Ai
h UA~
h A~L
a1U
a1L aU2 a 2L
a3L a3U a 4L
Fig. 1. The representation of interval type-2 fuzzy set.
aU4
x
3 Proposed Model
This section focuses on the development of fuzzy MCDM model that is incorporated
with intuitive multiple centroid for interval type-2 fuzzy sets.
Step 1: Determine the weights of evaluation criteria.
The weighting of evaluation criteria are employed.
Step 2: Construct a pairwise comparison matrices.
The pairwise comparison matrices are constructed among all criteria in the dimension
of the hierarchy systems based on the decision makers’ preferences using following
matrix:
a~12 a~1n
1 a~12 a~1n 1
a~
~
~
1
1 a2n 1/ a12
a~2n
(2)
A 21
~
~
~
~
an1 an 2 1 1/ a1n 1/ a2n 1
Step 3: Aggregate the decision makers’ preferences.
The pairwise comparison matrices of decision makers’ preferences are aggregated
using equation below:
~ (a
~1 a
~ 2 ... a
~ n )1 / k
a
ij
ij
ij
ij
where k is the number of decision makers and i=1,2,…m; j=1,2,…n.
(3)
Step 4: Defuzzify the fuzzy numbers of aggregation’s result of decision makers’
preferences.
The intuitive multiple centroid (IMC) defuzzification is extension from the classical
vectorial centroid method for fuzzy numbers that proposed by [9], [10]. The concept
is similar like the other centroid methods, to find the best centre point of fuzzy
numbers that represent in crisp values or single values. Comparing to other centroid
methods, IMC produces the appropriate way to get the output values that are more
intelligent manner, easy to compute, more balance and consider all possible cases of
fuzzy numbers. The IMC formula can be summarised as follows:
2(aU a1L aU4 a4L ) 7(aU2 a2L a3U a3L ) 7 U
, (h A~ hA~L )
IMC ( ~
x A~ , ~y A~ ) 1
A
36
36
(4)
x A~ , ~
y A~ ) with
Step 5: Compute the centroid index of intuitive multiple centroid of ( ~
vertices , , , and
,
.
Centroid index of intuitive multiple centroid can be generated using Euclidean
Distance by [11] as:
~
(5)
R ( A) ~
x 2 ~y 2
Step 6: Compute the criteria values as weightage for alternatives’ evaluation using
consistent fuzzy preference relations.
Consistent fuzzy preference relations was proposed by [12] for constructing the
decision matrices of pairwise comparisons based on additive transitivity property.
Referring to [13], a fuzzy preference relation R on the set of the criteria or
alternatives A is a fuzzy set stated on the Cartesian product set A A with the
membership function R : A A 0,1 . The preference relation is denoted by n n
matrix R (rij ) where rij y (ai , a j ) i, j 1,..., n . The preference ratio, rij of
the alternative ai to a j is determined by
ai is different to a j
0.5
rij (0.5,1) ai is preferred than a j
1
ai is absolutely preferred than a j
The preference matrix R
is presumed to be additive reciprocal,
pij p ji 1, i, j 1,..., n . Several propositions are associated to the consistent
additive preference relations as follows:
Proposition 1 [14]: Consider a set of criteria or alternatives, X x1 ,..., x n , and
associated with a reciprocal multiplicative preference relation
A (aij ) for
1
a ij ,9 . Then, the corresponding reciprocal fuzzy preference relation, P ( pij )
9
with pij 0,1 associated with A is given by the equation
1
pij g (aij ) (1 log 9 aij )
2
(6)
1
1
Generally, if a ij , n , then log n aij is used, in particular, when a ij ,9 ;
n
9
1
log 9 aij is considered as in the above proposition because aij is between and 9. If
9
1
aij is between and 7, then log 7 aij is used.
7
Proposition 2 [14]: For a reciprocal fuzzy preference relation P ( pij ) , the
following statements are equivalent.
(i)
(ii)
3
p ij p jk p ki , i, j , k
2
3
p ij p jk p ki , i j k
2
(7)
(8)
Proposition 3 [14]: For a reciprocal fuzzy preference relation P ( pij ) , the
following statements are equivalent
3
, i j k
2
(i)
p ij p jk p ki
(ii)
pi (i 1) p (i 1)(i 2) ... p ( j 1) j p ji
j i 1
, i j
2
(9)
Proposition 3 is crucial because it can be used to construct a consistent fuzzy
preference relations form the set of n 1 values p12 , p 23 ,..., p n 1 . A decision
matrix with entries that are not in the interval [0,1] , but in an interval c,1 c, c 0 ,
can be obtained by transforming the obtained values using a transformation function
that preserves reciprocity and additive consistency with the function
f : c,1 c 0,1 , f ( x)
( x c)
(1 2c)
(10)
Step 7: Ranking evaluation of alternatives using fuzzy TOPSIS
Concept of TOPSIS method originally proposed by [15]. They claim that the
alternative should not be chosen based on having the shortest distance from the
positive ideal reference point (PIRT) only, but also have the longest distance from the
negative ideal reference point (NIRP) in solving the MCDM problems. Here, the
extension of fuzzy TOPSIS is illustrated differs from others in terms of the usage of
defuzzification method, normalization process and ranking. The fuzzy decision matrix
is constructed and the linguistic terms from fuzzy numbers are used to evaluate the
alternatives with respect to criteria. Then, aggregate DMs’ preferences:
A1
A2
DM
Am
C2 Cn
~
x12 ~
x1n
~
x 22 a~2 n
~
~
x m1 x mn
(11)
1 ~1
xij , , ,~
xijk , , , ~
xijK
K
is the performance rating of alternatives, Ai with respect to criterion C i
i 1,2,..., m;
where xij
C1
x11
~
~
x 21
~
x m1
j 1,2,..., n
,~
xij
evaluated by kth experts and
~
xij (a1k , a 2k , a3k , a 4k ; h k ) . Fuzzy decision matrix is
weighted and normalised. Then, defuzzify the standardised generalised fuzzy numbers
into coordinate form, ( ~x , ~y ) . The weighted fuzzy normalised decision matrix is
~
denoted by V as depicted below:
~
; i 1,2,..., m;
V v~ij
mn
where
~
v~ij ~
xij w
j
j 1,2,..., n
(12)
(13)
Normalised each generalised trapezoidal fuzzy numbers into standardised generalised
fuzzy numbers. The weights from consistent fuzzy preference relations are adopted
here. Defuzzify the standardised generalised fuzzy numbers using intuitive multiple
centroid, then translate them into the index point. Use the new point of y A~ to
i
compute the index centroid point of standardised generalised trapezoidal fuzzy
2
2
~
numbers using Euclidean distance equation: R( Ai* ) ~xi ~y iS
Determine the
.
fuzzy positive-ideal solution (FPIS) and fuzzy negative-ideal solution (FNIS).
Referring to normalise trapezoidal fuzzy weights, the FPIS, A represents the
compromise solution while FNIS, A represents the worst possible solution. The
range belong to the closed interval [0,1]. The FPIS A (aspiration levels) and FNIS
A (worst levels) as following below:
A [1,1,1,1;1][1,1,1,1;0.9] A [1,1,1,1;1][1,1,1,1;0.9]
The FPIS, A and FNIS, A can be obtained by centroid method for ( x A , y A ) and
( x A , y A ) .
~
~
The distance d i and d i of each alternative from formulation A and A can be
calculated by the area of compensation method:
d i (v~ij , v~ j ) ( x A~* x A ) 2 ( y A~* y A ) 2
(14)
d i (v~ij , v~ j ) ( x A~* x A ) 2 ( y A~* y A ) 2
(15)
i
i
i
i
Find the closeness coefficient, CC i and improve alternatives for achieving aspiration
levels in each criteria. Notice that the highest CC i value is used to determine the
rank.
CC i
di
di di
1
di
(16)
di di
where,
di
di di
is satisfaction degree in ith alternative and
di
di di
is fuzzy gaps
degree in ith alternative.
Fuzzy gap should be improvised for reaching aspiration levels and get the best
mutually beneficial strategy from among a fuzzy set of feasible alternatives.
Step 8: Validation process using sensitivity analysis
The results of fuzzy MCDM models are importantly needed to validate using
sensitivity analysis method to analyse the quality and how robustness of fuzzy
MCDM model to reach a right decision under different conditions. In this paper,
sensitivity analysis that proposed by [6] is utilised for validation purposes.
4 Case Study
This section illustrates a numerical example for proposed hybrid fuzzy MCDM
methodology based on real case study for staff recruitment problem in MESSRS
SAPRUDIN, IDRIS & CO firm in Malaysia. The legal company plan to hire the best
candidate for executive post in several aspects which there are three decision makers
(DMs) DM1, DM2, and DM3 of a firm and four alternatives or candidates x1, x2, x3
and x4. Several criteria are considered which are: C1) Emotional steadiness, C2)
Oration, C3) Personality, C4) Past experience and, C5) Self-confidence. This study
simplifies the concept of attributes under fuzzy events. The values of attributes
correspond to interval type-2 fuzzy sets. A comparative study was conducted to
validate the results of the proposed model with established hybrid model which is
fuzzy AHP – fuzzy TOPSIS proposed by [16].
Table 1. Ranking results of criteria for comparing study.
Fuzzy AHP – TOPSIS [16]
ES
0.087
Criteria weight values
O
P
PE
0.364
0.044
0.34
S-C
0.164
O>PE>S-C>ES>P
Proposed Model
0.1172
0.2672
0.2219
PE>O>S-C>P>ES
Fuzzy MCDM model
0.1190
Ranking results
0.2747
Table 1 represents the criteria weight and ranking results between established fuzzy
AHP – fuzzy TOPSIS [16] and proposed model. Based on decision makers’
evaluation; Past experience, Oration and Self-confidence criteria play important
aspects in recruiting new staff since the weight are greater than 0.2 respectively.
These results of criteria’s weights are implemented in following phase to evaluate for
alternatives selection. The established fuzzy AHP – fuzzy TOPSIS model [16]
produces different ranking results with rank Oration > Past Experience > Selfconfidence > Emotion Steadiness > Personality.
Table 2. Ranking results of alternatives for comparing study
Fuzzy MCDM model
Alternatives ranking values, CCi
Alt1
Alt2
Alt3
Alt4
Ranking results
Fuzzy AHP – TOPSIS [16]
0.5497
0.5543
0.5616
0.5413
C3>C2>C1>C4
Proposed Model
0.7422
0.7823
0.83
0.6964
C3>C2>C1>C4
Table 2 depicts the alternatives/ candidates ranking results for CC i values. The
proposed model evaluates candidate 3 as the highest rank with 0.83 followed by
candidate 2, candidate 1 and candidate 4 for the last rank. The results reveal that the
candidate 3 is most suitable for this recruitment post. The established model [16]
produces same ranking results for alternatives with the proposed model. Even, this
model gives same ranking to proposed model, but the gaps of CCi values between
each candidate are too small. Those ranking results by [16] will easily affected if the
weightage of criteria are slightly changed. This can be evaluated by sensitivity
analysis in studying how consistent and robust the model. In the context of sensitivity
analysis evaluation, it presents that the proposed hybrid fuzzy MCDM model is
definitely consistent even the weights of criteria are changed with several
percentages. From the consistency results, the proposed hybrid fuzzy MCDM model
is recommended to deal with bigger case study in real world phenomena in order to
solve human based decision making problems under fuzzy environment.
Fig. 2. Sensitivity analysis results by varying the weights of the criteria by proposed
model.
Fig. 3. Sensitivity analysis results by varying the weights of the criteria by [16].
Fig. 2 illustrates the analysis results of changing the criteria weights for proposed
model. It presents that when the weights of the criteria change, the values of the CC i
vary slightly. As can be seen from Fig. 3, the values and patterns of changes of CC i
are too small compare to the proposed model. The ranking values between alternative
to other alternative are too small. That’s mean that the gap are small to represent the
assessment status of acceptance. This is depicted that the proposed model is good in
robustness than established model [16].
5 Conclusion
This study has brought out the idea and concept regarding the fuzzy MCDM model
that consist of consistent preference relations and fuzzy TOPSIS using intuitive
multiple centroid (IMC) defuzzification method based on interval type-2 fuzzy sets.
The development of IMC provides efficient computational defuzzification procedures
for fuzzy sets. It presents in simple formulae that based on the perspective of analytic
geometric principles. In developing an intuitionistic defuzzification, a novel manner
of computing intuitive multiple centroid method has capability in dealing with all
possible cases of interval type-2 fuzzy numbers. The development of fuzzy MCDM
model provides better selection in human based decision making problems where at
the same capable to deal with uncertainty in human judgment. Due to access
information and availability of the uncertain data, it is hard to make right decision. In
this sense, it is important to improvise the techniques or models form the classical
one, adding intuitive reasoning and human subjectivity. As consequence, the proposed
model is developed to design the robust and consistent methodology in order to give
the most promising alternatives with respect to the resources. Therefore, this proposed
model can be further proceeded in order to make some contributions by considering
complicated case studies drawn for diverse fields crossing human based decision
making problems.
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
G. J. Klir, U. St. Clair, and B. Yuan, Fuzzy set theory: foundations and applications, Internatio.
Prentice Hall, 1997.
C. Wagner and H. Hagras, “Uncertainty and type-2 fuzzy sets and systems,” Comput. Intell.
(UKCI), 2010 UK Work., pp. 1–5, 2010.
N. N. Karnik and J. M. Mendel, “Centroid of a type-2 fuzzy set,” Informayion Sci., vol. 132, pp.
195–220, 2001.
N. Mogharreban and L. F. Dilalla, “Comparison of defuzzification techniques for analysis of noninterval data,” Annu. Conf. North Am. Fuzzy Inf. Process. Soc. - NAFIPS, no. 1, pp. 257–260,
2006.
N. N. Karnik and J. M. Mendel, “Centroid of a type-2 fuzzy set,” Inf. Sci. (Ny)., vol. 132, no. 1–4,
pp. 195–220, Feb. 2001.
A. Amini and A. Alinezhad, “Sensitivity Analysis of TOPSIS Technique: The Results of Change
in the Weight of One Attribute on the Final Ranking of Alternatives,” J. Optim. Ind. Eng., vol. 7,
no. 2011, pp. 23–28, 2011.
H. Deng, “Comparing and ranking fuzzy numbers using ideal solutions,” Appl. Math. Model.,
vol. 38, no. 5–6, pp. 1638–1646, Mar. 2014.
Y. Gong, N. Hu, J. Zhang, G. Liu, and J. Deng, “Multi-attribute group decision making method
based on geometric Bonferroni mean operator of trapezoidal interval type-2 fuzzy numbers,”
Comput. Ind. Eng., vol. 81, pp. 167–176, 2015.
K. M. N. Ku Khalif and A. Gegov, “Generalised Fuzzy Bayesian Network with Adaptive
Vectorial Centroid,” in 16th world congress of the international fuzzy systems association (IFSA)
and the 9th conference of the European society for fuzzy logic and technology (EUSFLAT), 2015,
pp. 757–764.
K. M. N. Ku Khalif and A. Gegov, “Bayesian Logistic Regression using Vectorial Centroid for
Interval Type-2 Fuzzy Sets,” in 7th International Joint Conference on Computational Intelligence
(IJCCI 2015), 2015, pp. 69–79.
C.-H. Cheng, “A new approach for ranking fuzzy numbers by distance method,” Fuzzy Sets Syst.,
vol. 95, no. 3, pp. 307–317, May 1998.
E. Herrera-Viedma, F. Herrera, F. Chiclana, and M. Luque, “Some issues on consistency of fuzzy
preference relations,” Eur. J. Oper. Res., vol. 154, no. 1, pp. 98–109, 2004.
N. H. Kamis, K. Abdullah, H. Mohamed, S. Sudin, and W. Z. A. W. Ishak, “Decision making
models based on consistent fuzzy preference relations with different defuzzification methods,”
2011 IEEE Colloq. Humanit. Sci. Eng. CHUSER 2011, no. Chuser, pp. 845–850, 2011.
T. C. Wang and Y. H. Chen, “Applying consistent fuzzy preference relations to partnership
selection,” Omega, vol. 35, no. 4, pp. 384–388, 2007.
C.-L. Hwang and K. Yoon, Multiple Attribute Decision Making. New York: Springer, 1981.
M. Kiliç and İ. Kaya, “Investment project evaluation by a decision making methodology based on
type-2 fuzzy sets,” Appl. Soft Comput., vol. 27, no. 2015, pp. 399–410, 2015.