Computers & Industrial Engineering 122 (2018) 24–38

Contents lists available at ScienceDirect

Computers & Industrial Engineering

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

Heterogeneous multigranulation fuzzy rough set-based multiple attribute T

group decision making with heterogeneous preference information
⁎
Bingzhen Suna, Weimin Mab, , Xiangtang Chenb,c, Xiaonan Lid
a
School of Economics and Management, Xidian University, Xi’an, Shaanxi 710071 China
b
School of Economics and Management, Tongji University, Shanghai 200092, China
c
Business Administration Department, Wenzhou Vocational and Technical College, Wenzhou 325035, China
d
School of Mathematics and Statistics, Xidian University, Xi’an, Shaanxi 710071, China

A R T I C LE I N FO A B S T R A C T

Keywords: This study proposes a heterogeneous attribute multigranulation fuzzy rough set approach to the problem of
Multiple attribute group decision-making multiple attribute group decision-making (MAGDM) under uncertainty. We first present the definition of het-
(MAGDM) erogeneous attribute multi-source information systems and then construct the heterogeneous multigranulation
Multi-source heterogeneous attribute approximation space using the arbitrary binary relation classes generated by different categories attribute. We
information system
then give the rough approximation of a fuzzy decision-making object with respect to heterogeneous multi-
Rough set
granulation approximation space, i.e., the heterogeneous attribute multigranulation fuzzy rough set model over
Multigranulation fuzzy rough set
Heterogeneous attribute values multi-source information system. Meanwhile, we investigate some interesting properties and conclusions for the
proposed new model and also discuss the interrelationship between the proposed heterogeneous attribute
multigranulation fuzzy rough set model and the existing generalized rough set models. After that, we construct a
new approach to MAGDM problems based on heterogeneous attribute multigranulation fuzzy rough set theory.
The decision-making procedure and the methodology as well as the algorithm of the proposed method are given
and a detailed comparison of the traditional methods to MAGDM problems illustrates the advantages and lim-
itations. Finally, an example of handling MAGDM problem of evaluation of emergency plans for unconventional
emergency events illustrates this approach.

1. Introduction
We constantly make decisions in our private and professional life.
The basic problem of decision-making for any individual is how to
reach a final result from a given set of finite number of alternatives by
handling the objection, various criteria and evaluation information in
practice. There will be a group decision-making (GDM) problem when
more than one individual takes part in a considered decision-making
problem. For a GDM problem, experts in the group express their pre-
ference opinions (attitudes) on alternatives and interact to derive a
common solution. As a matter of fact, a MAGDM problem is described
as the selection of the best alternative among m alternatives while
trading-off between n attributes with the preference evaluation given
by all experts in the group. Then the complexity of the analysis in-
creases dramatically when moving from a single decision maker to a
multiple decision maker setting (Hwang & Lin, 1987; Ma, Zhan, Ali, &
Mehmood, 2018; Zhan & Alcantud, 2018). The problem no longer de-
pends on the preferences of a single decision maker; nor does it simply
involve the summing up of preferences of multiple decision makers
(Dong, Zhang, & Herrera-Viedma, 2016; Özgür & Bilal, 2017; Xu, Li, &
Wang, 2013; Xu, Cabrerizo, & Herrera-Viedma 2017). So, dealing with
the preferences given by experts in group and then reach a consensus
solution is the key issue of a MAGDM problem.
As an effective and powerful approach to the complexity decision-
making problems under uncertainty, the MAGDM methods are used to
deal with complexity decision-making problems with multiple decision
makers increases dramatically of reality. For a considered MAGDM
problem, the first step is all decision makers (or experts, stakeholders,
participants, etc.), which may have different backgrounds and knowl-
edge on the problem on hand, provide evaluations regarding to per-
formances of the alternatives under multiple criteria, and then obtain
the comprehensive result for all alternatives by fusion the evaluations
given by all decision makers. Since many multiple dimensional decision
problems of different fields requires multiple experts and/or decision
makers, MAGDM methods are receiving considerable interest in many
different research fields (Shen, Xu, & Xu, 2016) such as energy (Onar,
Oztaysi, Otay, & Kahraman, 2015), logistics (Kucukaltan, Irani, &
Aktas, 2016), safety management (Inan, Gul, & lmaz, 2017), facility

⁎
Corresponding author.
E-mail addresses: bzsun@xidian.edu.cn (B. Sun), mawm@tongji.edu.cn (W. Ma).

https://doi.org/10.1016/j.cie.2018.05.034
Received 25 September 2017; Received in revised form 19 May 2018; Accepted 21 May 2018
Available online 24 May 2018
0360-8352/ © 2018 Published by Elsevier Ltd.
B. Sun et al.
location (Chauhan & Singh, 2016), business process management
(Maio, Fenza, & Loia, 2016), and supplier selection and sustainable
development (Ünlüç, Ç Ervural, Ervural, & Kabak, 2017). Detailed re-
view for the studies of MAGDM which are refer to Özgür and Bilal
(2017). At the same time, many different mathematical theories such as
fuzzy set theory (Zadeh, 1997), interval mathematic (Zhou & Wu,
2008), probability theory and mathematical programming (Savage,
1954) are used to construct the model and method for all existing ap-
proaches to MAGDM problems.
Recently, a new mathematical theory of handling uncertainty
named as rough set (Pawlak, 1982) is used to MAGDM problems (Sun,
Ma & Qian, 2017; Sun, Ma & Xiao, 2017). The basic philosophical ideas
of Pawlak rough set comes from the perception that objects char-
acterized by the same description are indiscernible in light of the
available information about them. The equivalence classes or elemen-
tary building blocks which are induced from an indiscernibility relation
form a partition of the universe of discourse, and constitute the basic
granules of knowledge (Sun, Ma & Qian, 2017; Sun, Ma & Xiao, 2017).
By using the concept of lower and upper approximations in rough set
theory, knowledge hidden in information systems may be unraveled
and expressed in the form of decision rules. Though the equivalence or
indiscernibility relation provides a powerful tool to deal with the ap-
proximate the inaccurate and uncertain target in decision information
systems, the partition or indiscernibility relation also is restrictive for
many applications. To overcome such unreasonableness, several ex-
tension forms of Pawlak rough set models have established. In general,
the relation-oriented extensions (i.e., relaxing the restriction of the
equivalence relation of the universe of discourse) (Slowinski &
Vanderpooten, 1990; Wang, Shao, He, Qian, & Qi, 2016; Wang, Hu,
Wang, Chen, & Qian, 2017; Wang, He, Shao, & Hu, 2017; Yao, Wong, &
Wang, 1995) and decision-making object-oriented extensions (i.e.,
combing the other mathematical theories such as fuzzy set theory, Dempster-
Shafer theory of evidence, and intuitionistic fuzzy set theory) (Chakhar,
Ishizaka, Labib, & Saad, 2016; Sun, Ma, & Zhao, 2016; Sun, Ma, & Chen,
2014) and other related problems (Kong, Zhang, & Ye, 2016; Ma et al.,
2018) are two mainly directions of improving the classical Pawlak
rough set model. Nowadays, the classical Pawlak rough set and its ex-
tensions have become important and efficiently theory and tool to deal
with various decision-making problems under uncertainty. Meanwhile,
several models and methodologies for decision-making under un-
certainty are proposed based on Pawlak rough set and its extensions
(Abastante, Bottero, Greco, & Lami, 2014; Chakhar et al., 2016;
Ishizaka & Nemery, 2013, 2013).
Granular computing, established by Zadeh (1997), has attracted
many researchers and practitioners as a new and rapidly growing
paradigm of information processing. Granular computing is referred an
umbrella term to cover several theories, methodologies, techniques,
and tools that make use of information granules in complex problem
solving (Yao, Vasilakos, & Pedrycz, 2013). From the perspective of
granular computing (Zadeh, 1997), an indiscernibility relation on the
universe of discourse can be regarded as a granularity, and the corre-
sponding partition can be regarded as a granular structure (Yao et al.,
2013). Then, the Pawlak rough set theory is based on a single granu-
larity (only one indiscernibility or equivalence relation). However,
there may be several granular structures when the rough set theory is
using to handle decision-making problems under uncertainty. So, ex-
tending the single granularity to multiple granularity structures for the
classical Pawlak rough set theory is necessity. Inspired by the concept of
granulation computing, Qian, Liang, Yao, and Dang (2010) defined the
multigranulation rough set model, where the set approximations were
defined using multiple equivalence relations on the universe of dis-
course, by replacing the single granulation with multiple granulation on
the universe of discourse. Then, under the framework of multiple
granularity structures, several generalized multigranulation rough set
models have established with various backgrounds of decision-making
under uncertainty. Detailed review for multigranulation rough set
25
Computers & Industrial Engineering 122 (2018) 24–38
theory is suggested refer to Sun, Ma, Li, and Li (2018).
It is easy to know that the attribute values in the existing multi-
granulation rough set models are usually the same type such as fuzzy
attribute values, symbol values, intuitionistic fuzzy set, or interval-va-
lued attribute values (Sun et al., 2018). In many applications, especially
management decision-making in practice, however, there may not only
be different combination of the selected evaluation attributes but also
different categories for the attribute values which describe the objects
or alternatives of a universe of discourse (Ma et al., 2018). We consider
a MAGDM problem of reality, all invited experts may come from dif-
ferent areas with different background and knowledge and then present
the evaluation for all candidate alternatives. On the one hand, the ex-
perts usually select different evaluation indices from all criterion which
are familiarized with themselves background and knowledge or their
personal preference. That is, different experts may select different
evaluation indices as the optimal combination to express their pre-
ference evaluation for all candidate alternatives. So, the preference
evaluations related to the optimal combination of the selected evalua-
tion indices given by different experts are made of a multiple granu-
larity structure of all candidate alternatives. On the other hand, because
the experts come from different areas and then the evaluation results
given by different experts will be natural numbers between 0 and 100,
and it can also be graded as Excellent, Good, Moderate, Bad, and Un-
acceptable. Sometimes, if needed, it might be graded into two values,
Accept, and Reject. Furthermore, there are other values could be given
such as interval-value, and vague-values or fuzzy-values. So, the eva-
luation results given by the experts for all candidate alternatives are
heterogeneous values with respect to different indices. Therefore, the
MAGDM problem is a multiple granularity and heterogeneous attribute
values structure with many different decision-makers (experts). Then,
how can obtain the final optimal alternative from all candidates under
the conditions of different decision-makers that select differently eva-
luation indices to express their preference evaluation with hetero-
geneous attribute values? For this reason, the existing rough set models
with one single granularity and multiple granularity with the same type
of available information for all indices (attributes) are incapable of
handling this type of decision-making problem. Therefore, a new gen-
eralization model of the existing multigranulation rough set (Qian
et al., 2010; Wu & Leung, 2011) should be defined and then establishes
a new perspective and theoretical tool to deal with the above problems
of decision-making under uncertainty. That is, the multigranulation
rough set model based on heterogeneous attribute values and the cor-
responded approach to decision-making problem under uncertainty.
Based on the above description for the decision-making problem of
selecting supplier and the philosophical ideas of existing multi-
granulation rough set theory with same type of attribute values
(Ishizaka & Nemery, 2013, 2013; Qian et al., 2010), the objective of this
paper is to propose an approach to MAGDM problems with hetero-
geneous values by defining a new multigranulation fuzzy rough set
model over heterogeneous attribute information systems. We first pre-
sent the basic definition of the multigranulation fuzzy rough set model
over heterogeneous attribute information systems, i.e., heterogeneous
attribute multigranulation rough set model. We then investigate the
properties and the relationship between the established new multi-
granulation model with the existing generalized rough set models
(Ishizaka & Nemery, 2013, 2013; Qian et al., 2010; Sun et al., 2018; Wu
& Leung, 2011) in detail. Also, some interesting conclusions are given
for the proposed heterogeneous attribute multigranulation fuzzy rough
set model. Finally, we present the methodology for MAGDM problems
with heterogeneous values based on heterogeneous attribute multi-
granulation fuzzy rough set model. Meanwhile, the detailed procedure
of decision-making for the proposed model is given. Meanwhile, the
approach is illustrated in detail through an example of the optimal
selecting supplier in the decision-making of supply chain.
The rest of this paper is organized as follows. Section 2 briefly re-
views the fuzzy set theory, Pawlak rough set and multigranulation
B. Sun et al.
rough set model with the properties. In Section 3, we present the multi-
source heterogeneous attribute information system and present the
characteristic function for every binary relation classes generated by
different values of attribute. Then we present the rough approximation
of a crisp (fuzzy) decision-making object of the universe with respect to
heterogeneous attribute multi-source information system. That is, the
heterogeneous multigranulation rough fuzzy set (fuzzy rough set)
model over multi-source heterogeneous attribute decision information
system. In Section 4, we consider a kind of MAGDM problem with
heterogeneous attribute values and then present a new decision model
and method by using the heterogeneous attribute multigranulation
fuzzy rough set theory. In Section 5, an illustrate example with the
background of decision-making problem of evaluation of emergency
plans for unconventional emergency events is used to test the efficiency
and validity of the proposed decision-making model and method. At
last, we conclude our research and set out further research directions in
Section 6.
2. Preliminaries
In this section, we briefly review the concept of the fuzzy set (Zadeh,
1997), Pawlak rough set (Pawlak, 1982) and multigranulation rough set
and its basic operations (Qian et al., 2010).
2.1. Fuzzy set
As is well known, the concept of fuzzy set (Zadeh, 1997) is a gen-
eralization of the classical crisp set (Yan, 1998). The membership
function of a fuzzy set is the naturally generalization of the concept of
characteristic function of classical crisp set. In the following, we present
the definition of fuzzy set by using the concept of characteristic func-
tion of classical crisp set.
Let U be a non-empty finite universe. For any A ⊆ U , we introduce a
characteristic function A (x ) as follows:
1, x ∈ A;
A (x ) = 1A (x ) = ⎧
⎩ 0, x ∉ A.
⎨ relation P and Q if X P + Q = X P + Q ; Otherwise, X is a rough set with
The characteristic function A (x ) is a mapping from U to {0, 1} . Then
any characteristic function on universe U determines a classical subset
of U. That is, A = {x ∈ U |A (x ) = 1} .
Next, we present the concept of fuzzy set by using the definition of
classical crisp set.
Definition 2.1 (Zadeh, 1997). Let U be a non-empty finite universe. A
fuzzy set of universe U is defined by the mapping A (•): U ⟶ [0, 1],
where the A (x ) denotes the membership of the element x (x ∈ U ) with
fuzzy set A.Let F (U ) stands for all fuzzy subsets of universe U.
Meanwhile, it is easy to see that the membership function A (•) will
be degenerated into the characteristic function when the domain of
[0, 1] becomes {0, 1} .For any A ∈ F (U ) , the r level set and strong r level
set of A will be denoted by Ar and Ar +, respectively. That is,
Ar = {x ∈ U |A (x ) ⩾ r } and Ar + = {x ∈ U |A (x ) > r } , where r ∈ [0, 1],
the unit interval, A0 = U and A1 + = ∅.
2.2. Pawlak rough set
Generally speaking, Pawlak rough set begins from the binary
equivalence relation of universe of discourse. Then the lower and upper
approximations of any target set are given by the equivalence classes
generated by the binary equivalence relation.
Let U be a non-empty finite universe and R be an equivalence re-
lation of U × U . The equivalence relation R induces a partition of U,
denoted by [x ]R or [x ], and U / R = {[x ]|x ∈ U } stands for the equivalence
classes of x. Then (U , R) is the Pawlak approximation space.
For any X ⊆ U , its lower and upper approximations are defined as
26
Computers & Industrial Engineering 122 (2018) 24–38
follows [V ]:
R (X ) = {x ∈ U |[x ] ⊆ X } = ∪{[x ]|[x ] ⊆ X },
R (X ) = {x ∈ U |[x ] ∩ X ≠ ∅} = ∪{[x ]|[x ] ∩ X ≠ ∅}.
The lower approximation R (X ) is the union of all elementary sets
which are the subset of X, and the upper approximation R (X ) is the
union of all elementary sets which have a non-empty intersection with
X. The positive, boundary and negative regions of X can be defined as
follows (Pawlak, 1982):
pos (X ) = R (X ), bn (X ) = R (X )− R (X ), neg (X ) = U −R (X ).
The positive region pos (X ) consists of all objects that are definitely
contained in the set X. The negative region neg (X ) consists of all objects
that are definitely not contained in the set X. The boundary region
bn (X ) consists of all objects that may be contained in X. Because ap-
proximations are from equivalence classes, inclusion into the boundary
region reflects uncertainty about the classification of objects.
2.3. Multigranulation rough set
We now present the definition of multigranulation rough set model
established by Qian et al. (2010).
Definition 2.2 (Qian et al., 2010). Let K = (U , R) be a knowledge base
and P , Q be two equivalence relations of universe U. For any X ⊆ U , the
lower and upper approximations of X with respect to P and Q are
defined as follows, respectively.
X P + Q = {x ∈ U |[x ]P ⊆ X ∨ [x ]Q ⊆ X },
X P + Q = ( X c P + Q )c.
where X c stands for the complementary of X.Meanwhile, the boundary
of X with respect to the equivalence relation P and Q is defined as:
BnP + Q = X P + Q − X P + Q .
Generally speaking, we call X definable with respect to equivalence
respect to K = (U , R) . Moreover, it is easy to know that any one
equivalence relation will form a partition of universe U, i.e., any one
equivalence relation of the universe forms a granularity structure of the
universe of discourse. So, we call X the multigranulation rough set
when X P + Q ≠ X P + Q because the granularity structure of the universe
U is generated by two different equivalence relations P and Q.The
definition of the lower and upper approximations of the
multigranulation rough set model could describe the decision process
well for decision-making about risk.
Proposition 2.1 (Qian et al., 2010). Let K = (U , R) be a knowledge base
and P , Q be two equivalence relations of universe U. For any X ⊆ U . Then
there are
X P+Q ⊆ X P∪Q and X P+Q ⊇ X P∪Q ,
where X P ∪ Q and X P ∪ Q denote the lower and upper approximations of
Pawlak rough set.Proposition 2.1 reveals the interrelationship between the
Pawlak rough set and the multigranulation rough set. The difference between
Pawlak rough lower and upper approximations and multigranulation lower
and upper approximations is caused by two different approximation
methods.
3. Heterogeneous multigranulation rough fuzzy set over multi-
source information system
In this section, we first present the description of a multi-source
heterogeneous attribute information system. Then, we will define the
multigranulation rough fuzzy set model over multi-source information
system. Meanwhile, some interesting properties as well as the
B. Sun et al. Computers & Industrial Engineering 122 (2018) 24–38

Table 1 decision-making problems. However, the above decision-making pro-

A multi-source heterogeneous attribute information system of credit card ap- blem of credit card applicant includes several different types attribute
plicant. values and the complexity will be increased sharply when using the
U d1 d2 ⋯ dl method and model in Sun et al. (2014). So, exploring new approaches
to multi-source heterogeneous attribute information system is more
c1 c2 ⋯ cm c1 c2 ⋯ cm c1 c2 ⋯ cm c1 c2 ⋯ cm valuable and necessity, this paper will present a new way to deal with
⋯ ⋯ ⋯ ⋯ the multi-source heterogeneous attribute information system under the
x1 1 3 2 s2 s−3 s0 0.8 0.1 0.2 Y Y Y
x2 2 5 ⋯ 2 s−1 s4 ⋯ s3 0.3 0.5 ⋯ 0.2 Y N ⋯ Y framework of multigranulation rough set theory.
x3 5 1 ⋯ 5 s1 s0 ⋯ s2 0.2 0.1 ⋯ 0.6 N N ⋯ N In this paper, we confine there includes three different attribute
x4 3 3 ⋯ 4 s0 s−2 ⋯ s1 0.6 0.3 ⋯ 0.5 N N ⋯ N values of linguistic value, fuzzy value and logic value (or symbol value)
x5 4 4 ⋯ 3 s3 s1 ⋯ s2 0.4 0.4 ⋯ 0.3 Y Y ⋯ Y in the considered multi-source heterogeneous attribute information
x6 2 3 ⋯ 5 s−2 s3 ⋯ s4 0.2 0.3 ⋯ 0.5 Y Y ⋯ Y
x7 3 3 ⋯ 1 s1 s2 ⋯ s3 0.3 0.3 ⋯ 0.6 Y N ⋯ N
system. Generally speaking, more than three types of attribute values
x8 4 4 ⋯ 3 s3 s2 ⋯ s−2 0.3 0.4 ⋯ 0.3 N N ⋯ N can be discussed with the same way except the scale and the complexity
x9 2 2 ⋯ 4 s−2 s0 ⋯ s3 0.3 0.2 ⋯ 0.4 Y Y ⋯ Y of the considered decision-making problem.

relationship between the new defined model and the existing models 3.2. The λ -heterogeneous binary relation over multi-source heterogeneous
will also be discussed in detail. attribute information system

From the basic structure of the multi-source heterogeneous attribute

3.1. Multi-source heterogeneous attribute information system
We call four-tuple (U , C , D , ) a multi-source heterogeneous attri-
bute information system, where U = {x1, x2 , …, x n} is the universe of
discourse (i.e., the objects set or the alternatives set), C = {c1, c2, …, cm}
is the attribute set. ω = (ω1, ω2, …, ωm)T is the weight vector of attributes
m incapable to deal with the multi-source heterogeneous attribute in-
where ωh ⩾ 0 and satisfies ∑h = 1 ωh = 1. D = {d1, d2, …, dl} is the
set of source of information collection.  = {fk (x i ,cj )|i =
1, 2, …, n;j = 1, 2, …, m;k = 1, 2, …, l} is a family of mapping sets be-
tween universe U and attribute set C, where
fk : U × C → Vk (k = 1, 2, …, l) and Vk (k = 1, 2, …, l) is the domain of
attribute set C with the available information dk (dk ∈ D) . That is,
∀ x i ∈ U , cj ∈ C , dk ∈ D , fk (x i , cj ) ∈ Vk , i.e., fk (x i , cj ) is the evaluation
of the alternative x i with respect to the attribute cj according to the
available information dk .
In order to illustrate the above definitions and notations, we present
a multi-source heterogeneous attribute information system (U , C , D , )
with the background of an evaluation problem of a credit card appli-
cant. Suppose that U = {x1, x2 , …, x 9} is a set of nine applicants. There
are c1−Education, c2−Age , c3−Salary, …, cm−Creditrecord . Meanwhile,
the corresponding weight of every attribute is ω1, ω2 , …, ωm .
D = {d1, d2, …, dl} is l experts which present the evaluation for every
applicant with respect to the considered attributes according to their
preference information such as ranking by 1, 2, …, 5; or by the linguistic
values S = {s−4 = extremely poor , s−3 = very poor , s−2 = poor , s−1 =
slightly poor , s0 = fair , s1 = slightly good , s2 = good , s3 = very good ,
s4 = extremely good} ; or by fuzzy values; or by the logic values Yes and
No, etc. Table 1. presents the evaluation results given by different ex-
perts.
It is easy to see that there are multiple different type values for the
attributes given by all experts, i.e., the evaluation of the credit card
applicant is a multi-source heterogeneous attribute information system.
At the same time, the problem of decision-making for this credit card
applicant is how to ranking all applicants according to the variety types
evaluation information (i.e., the symbol values evaluation; the lin-
guistic values evaluation and logic values evaluation, ect.) given by
different experts. That is, what the comprehension evaluation for all
applicants with respect to the attributes by the fusion of the hetero-
geneous attribute values given by different experts.
In fact, Sun et al. (2014) has discussed two types hybrid attribute
values decision information systems and presented a fuzzy rough set
model by constructed a binary equivalence relation over the hybrid
attribute values decision information systems according to the original
fuzzy rough set theory (Slowinski & Vanderpooten, 1990). The fuzzy
rough set method given in Ref. (Sun et al., 2014) makes an initial ex-
ploring to hybrid attribute values decision information system and also
provides an effectively way to handle with hybrid attribute values
information system, we can easy to know that there will be more than
one binary relation generated by different sources of available in-
formation. So, the existing rough set models based on single binary
relation (Chen, Zhang, Zhao, Hu, & Zhu, 2012; Dubois & Prade, 1990;
Pawlak, 1982; Slowinski & Vanderpooten, 1990; Yao et al., 1995) are
formation system. Because multiple binary relations will be formed,
from the point of view of granular computing, there are forming a
multiple granulation structure over the multi-source heterogeneous
attribute information system. Then, the rough approximation of any
target objects will be discussed under the framework of multiple
granulation. This section will present the definition of heterogeneous
multigranulation approximation space for the multi-source hetero-
geneous attribute information system, and then present the generalized
rough set model over the heterogeneous multigranulation approxima-
tion space in detail.
In what follows, we first discuss how to build a heterogeneous
multigranulation approximation space over multi-source heterogeneous
attribute information system by constructing the binary relation classes
of any objects for the linguistic value attribute, fuzzy value attribute
and logic value (or symbol value) attribute, respectively.
According the assumption of Section 3.1, we reformulate the multi-
source heterogeneous attribute information system (U , C , D , ) as fol-
lows.
Let (U , C , D , ) be a multi-source heterogeneous attribute in-
formation system, where U is the universe of discourse, C is the attri-
bute set, and D is the set of source of information collection.
 = {fkΓ (x i , cj )|i = 1, 2, …, n;j = 1, 2, …, m;Γ = F , L, S;k = 1, 2, …, l} is a
family of mapping sets between universe U and attribute set C with the
weight vector ω = (ω1, ω2, …, ωm)T , where fkF , fkL and fkS stand for the
fuzzy value attribute, linguistic value attribute and symbol value at-
tribute obtained from available information dk (dk ∈ D) of the multi-
source heterogeneous attribute information system, respectively.
Subsequently, we present the definition of fuzzy similarity relation
over the fuzzy value attribute (Huseyin & Warren, 2005) for the multi-
source heterogeneous attribute information system (U , C , D , ) .
Definition 3.1. Let (U , C , D , ) be a multi-source heterogeneous
attribute information system. For any B ⊆ C and the weight vector
ω = (ω1, ω2, …, ωm)T , RkF (B ) is the fuzzy equivalence relation (or fuzzy
similarity relation) with the available information
dk (dk ∈ D , k ∈ {1, 2, …, l}) on B if and only if the following condition
holds:
RkF (B )(x i , x j ) = {α|(x i , x j ) ∈ U × U , α = min ωt ct (x i , x j ), ct ∈ C , i, j
ct ∈ B
= 1, …, n;1 < t < m;k ∈ {1, 2, …, l}} (1)

27
B. Sun et al. Computers & Industrial Engineering 122 (2018) 24–38

where ct (x i , x j ) is the degree of similarity of two objects x i , x j on (Yao et al., 1995). If sα ∈ S , we call sα the original term and α the
ct (1 < t < m) . So, α is called similarity degree on fuzzy attribute set B of original term index; otherwise, we call sα the virtual term and α the
x i , x j , denoted by rij = α = RkF (B )(x i , x j ) and rij ∈ [0, 1]. virtual term index. Specifically, the decision-maker uses the original
In general, a definition of similarity degree on a fuzzy value attri- linguistic terms to evaluate alternatives, and the virtual linguistic terms
bute set can be presented as follows (Sun et al., 2014): can only appear in operations.
Given a term set S , for any sα , sβ ∈ S , and μ, μ1 , μ 2 ∈ [0, 1], the
⎧1, i = j;
following operational laws hold (Dong et al., 2016; Xu et al., 2013):
Tk (x i , x j ) = min{ω (1−|f F (x , c )−f F (x , c )|)}, i ≠ j.
⎨ c ∈C t k i t k j t
⎩ t (2)
(1) sα ⊕ sβ = sα + β ;
Remark 3.1 (Huseyin and Warren, 2005). Based on the Definition 3.1, (2) μsα = sμα ;
the fuzzy equivalence relation RkF (B ) can be represented by a fuzzy (3) (μ1 + μ 2 ) sα = μ1 sα ⊕ μ 2 sα .
similarity matrix. In particular, if B = ∅, we also present the definition
of RkF (∅) as follows: For the convenience of calculation, we present a mapping between
r r12 ⋯ r1n any linguistic term set to a real number interval as follows.
⎛ 11 ⎞ ⎛1 ⋯ 1⎞
r21 r22 ⋯ r2n
RkF (B ) = ⎜ ⋯ ⋯ ⋯ ⋯ ⎟⎟, and RkF (∅) = ⋮ ⋱ ⋮ . Definition 3.3. Let S = {sα |s−q ⩽ sα ⩽ sq, α ∈ [−q, q]} be a continuous
⎜ ⎜ ⎟
⎝ rn1 rn2 ⋯ rnn ⎠ ⎝1 ⋯ 1⎠ linguistic values term set, we define

This is because we cannot distinguish any objects of a universe when I : S ⟶ [−q, q], i. e . , I (sα ) = α, for any sα ∈ S and α ∈ [−q, q].
the fuzzy attribute set is empty (Huseyin & Warren, 2005).Based on the
fuzzy equivalence relation of the universe of discourse, we can obtain Definition 3.4. Let (U , C , D , ) be a multi-source heterogeneous
the concept of fuzzy equivalence class induced by a fuzzy equivalence attribute information system. For any objects x i , x j ∈ U , the distance
relation. measure between x i and x j with respect to attribute
ct ∈ C (t = 1, 2, …, m) and the weight vector ω = (ω1, ω2, …, ωm)T based
Definition 3.2 (Zadeh, 1971). Let U be a non-empty finite universe. R is on the available information dk (dk ∈ D , k ∈ {1, 2, …, l}) is defined as
a fuzzy binary relation of universe U. We define a fuzzy class [x i]R of any
x i ∈ U induced by R as follows: |I (fkL (x i , ct ))−I (fkL (x j , ct ))|
ξckt (x i , x j ) = .
ri1 r r 2q (5)
[x i]R = + i2 + ⋯+ in , n = |U |.
x1 x2 xn (3)
It is clear that [x i]R is a fuzzy set of universe U. Furthermore, we call
By the fuzzy similarity matrix on the multi-source heterogeneous m
attribute information system (U , C , D , ) and the fuzzy similarity ξCk (x i , x j ) = ∑ ωt ξckt (x i , x j )
classes given in Definition 3.2, we now present the λ fuzzy similarity t=1 (6)
relation classes for any object x (x ∈ U ) as follows:
Let (U , C , D , ) be a multi-source heterogeneous attribute in- the distance measure between objects x i and x j of universe U with
formation system. RF is a binary fuzzy similarity relation of attribute set respect to the attribute set C based on the available information
C and [x i]k F is the fuzzy similarity classes over (U , C , D , ), respec- dk (dk ∈ D , k ∈ {1, 2, …, l}) .
R According to the definition of the distance measure given in (5) and
tively. For any λ (0 ⩽ λ ≤ 1) , we define
(6), one can easy to know there are 0 ⩽ ξckt ≤ 1 and 0 ⩽ ξCk ≤ 1 and both
[x i]kλ
RF
= {x j ∈ U |rijk ⩾ 1−λ, λ ∈ [0, 1], j = 1, 2, …, n, k the distance measure ξckt and ξCk are n × n symmetric matrix over the
universe U. Then, we present the definition of λ binary linguistic value
∈ {1, 2, …, l}}, xi ∈ U (4)
relation RL of any objects over universe based on the distance measure.
the λ binary relation classes of alternative x i ∈ U with the available
Definition 3.5. Let (U , C , D , ) be a multi-source heterogeneous
information dk (dk ∈ D , k = 1, 2, …, l) over (U , C , D , ). attribute information system. ξCk is the distance measure between
Next, we discuss the binary relation classes for the linguistic value objects x i and x j of universe U with respect to the attribute set C
attribute over multi-source heterogeneous attribute information system based on the available information dk (dk ∈ D , k ∈ {1, 2, …, l}) . We
(U , C , D , ) . define
The linguistic approach is an approximate technique, which re-
presents qualitative aspects as linguistic values by means of linguistic [x i]kλL = {x j ∈ U |ξCk (x i , x j ) ⩽ λ, λ ∈ [0, 1], j = 1, 2, …, n, k
R
variables (Dong, Li, & Herrera, 2016). In general, the linguistic vari-
ables are formally defined as follows. ∈ {1, 2, …, l}}, xi ∈ U , (7)
Let S = {sα |α = t , …, −1, 0, 1, …, t } be a finite and totally ordered
the λ binary linguistic value relation classes of object x i ∈ U over
discrete term set with odd cardinality, where sα represents a possible
(U , C , D , ) .
value for a linguistic variable. Specifically, s−t and st are the lower and
Finally, we discuss the binary relation classes for the symbol value
upper limits, respectively, and t is a positive integer. Furthermore, the
attribute over multi-source heterogeneous attribute information system
following conditions must be hold for the linguistic terms (Dong et al.,
(U , C , D , ) .
2016; Xu et al., 2013):
As is well known, there is a binary equivalence relation on the at-
tribute with symbol values of any information system (Pawlak, 1982;
(1) The set is ordered: sα ⩾ sβ if and only if α ⩾ β ;
Sun et al., 2018). So, we define the binary relation classes over multi-
(2) There is a negation operator: neg(sa )= s−α , specifically, neg(s0 )= s0 .
source heterogeneous attribute information system (U , C , D , ) as fol-
lows.
In general, the results of the modelling process for decision-making
preference information do not match the elements in S. To facilitate Definition 3.6. Let (U , C , D , ) be a multi-source heterogeneous
computation and preserve all the given information, the discrete term attribute information system. For any objects x i , x j ∈ U and
set S is extended to a continuous term set ct ∈ C (t = 1, 2, …, m) with the weight vector ω = (ω1, ω2, …, ωm)T the
S = {sα |s−q ⩽ sα ⩽ sq, α ∈ [−q, q]} (where q (q > t ) is a sufficiently large similarity measure of objects x i and x j based on the available
positive integer) whose elements also meet all the characteristics above information dk (dk ∈ D , k ∈ {1, 2, …, l}) is defined as

28
B. Sun et al. Computers & Industrial Engineering 122 (2018) 24–38

Table 2 mC (x1, x 4 ) = 0.2 × 1 + 0.3 × 0 + 0.4 × 0 + 0.1 × 0 = 0.2.

A multi-source heterogeneous attribute information system of credit card ap-
plicant. This is the result of Definition 3.6.
This completes the example.
U d1 d2 d3
So far we have established the λ binary relation classes for different
c1 c2 c3 c4 c1 c2 c3 c4 c1 c2 c3 c4 attribute values over multi-source heterogeneous attribute information
x1 2 2 3 1 s2 s−3 s1 s0 0.8 0.1 0.6 0.2
system (U , C , D , ) , respectively. Now, we define the λ heterogeneous
x2 1 2 3 2 s−1 s4 s−1 s3 0.3 0.5 0.3 0.2 binary relation over multi-source heterogeneous attribute information
x3 3 1 1 1 s1 s0 s2 s2 0.2 0.1 0.7 0.6 system (U , C , D , ) based on the above three types binary relation
x4 2 3 2 3 s0 s−2 s0 s1 0.6 0.3 0.5 0.5 classes.
Definition 3.7. Let (U , C , D , ) be a multi-source heterogeneous
m
attribute information system. For any x i ∈ U , the λ heterogeneous
mCk (x i , x j ) = ∑ ωt mckt (x i , x j ); where mckt (x i , x j ) binary relation classes R RλH is defined as
t=1

1, fkS (x i , ct ) = fkS (x j , ct ); [x i]λ H = {y ∈ [x i]kλΓ | ∀ y ∈ U ; Γ = F , L, S; k = 1, 2, …, l; λ ∈ [0, 1]}.

=⎧ R R
⎨
⎩ 0, Others. (9)
Obviously, the similarity measure on the symbol values attribute set Based on Definition 3.7, the following results are clear for the λ
determines an equivalence relation RSk over multi-source heterogeneous heterogeneous binary relation over multi-source heterogeneous
attribute information system (U , C , D , ) . Then, the λ binary symbol attribute information system (U , C , D , ) .
value relation classes of object x i ∈ U over (U , C , D , ) is as follows:
Proposition 3.1. Let (U , C , D , ) be a multi-source heterogeneous
[x i]kλS = {x j ∈ U |mCk (x i , x j ) ⩾ 1−λ, λ ∈ [0, 1], j = 1, 2, …, n, k attribute information system. RH is a heterogeneous binary relation of
R
attribute set C. Then the following properties are satisfied:
∈ {1, 2, …, l}}. (8)
In order to illustrate the idea and the process of computing for the (1) For any x i ∈ U , there is x i ∈ [x i]Rλ H (0 ⩽ λ ≤ 1) , i.e., [x i]Rλ H is reflex-
above three types λ binary relation classes for different attribute values ivity;
over multi-source heterogeneous attribute information system (2) For any x i , x j ∈ U , there is x j ∈ [x i]Rλ H ⇔ x i ∈ [x j]kλ
RH
(0 ⩽ λ ≤ 1) , i.e.,
(U , C , D , ) , we present a numerical example as follows. [x i]λ H is symmetry.
R
Example 3.1. Considering a multi-source heterogeneous attribute Proof. It can be directly derived from the above definitions. □
information system (U , C , D , ) as Table 2.
From the definition of the λ heterogeneous binary relation classes,
Suppose the weights of every attribute are as follows:
for any x i , x j ∈ U and satisfy x i ∈ [x j]Rλ H , there exists x 0 ∈ U satisfies
ω1 = 0.2, ω2 = 0.3, ω3 = 0.4, ω1 = 0.1.
x 0 ∈ [x i]Rλ H and x 0 ∉ [x j]Rλ H . That is, the λ heterogeneous binary rela-
Then, using the formula (1) and (2), we have tion class doesn’t satisfy the transitivity. So, the λ heterogeneous binary
r11 = 1, relation class is not an equivalence relation of the universe of discourse,
i.e., λ heterogeneous general binary relation of universe of discourse. At
r12 = min{0.2 × (1−|0.8−0.3|), 0.3 × (1−|0.1−0.5|), 0.4 × (1−|0.6 the same time, the following results are also clear for the λ hetero-
ci ∈ C
geneous binary relation.
−0.3|), 0.1 × (1−|0.2−0.2|)} = 0.1,
Proposition 3.2. Let (U , C , D , ) be a multi-source heterogeneous
r13 = min{0.2 × (1−|0.8−0.2|), 0.3 × (1−|0.1−0.1|), 0.4 × (1−|0.6 attribute information system. [x i]Rλ H (0 ⩽ λ ≤ 1) is the λ heterogeneous
ci ∈ C binary relation classes of universe U. Then there are
−0.7|), 0.1 × (1−|0.2−0.6|)} = 0.06,
(1) For any x i ∈ U , there is x i ∈ [x i]0 H ;
R
r14 = min{0.2 × (1−|0.8−0.6|), 0.3 × (1−|0.1−0.3|), 0.4 × (1−|0.6 (2) For any x i ∈ U , there is [x i]1R H = U ;
ci ∈ C
(3) For any x i ∈ U , λ1, λ2 ∈ [0, 1] and satisfies λ1 ⩽ λ2 , there is
−0.5|), 0.1 × (1−|0.2−0.5|)} = 0.07.
[x i]λ1H ⊆ [x i]λ2H .
R R
This is the result of Definition 3.1.
Proof. It can be directly verified by Definition 3.2, Definitions 3.5 and
Using the formula (5) and (6), we have.
3.6. □
ξC (x1, x1) = 0,
Furthermore, we can easily define the characteristic function of any
3 7 2 3 λ heterogeneous binary relation classes over multi-source hetero-
ξC (x1, x2) = 0.2 × + 0.3 × + 0.4 × + 0.1 × = 0.475,
8 8 8 8 geneous attribute information system (U , C , D , ) in the following.

1 3 1 2 Definition 3.8. Let (U , C , D , ) be a multi-source heterogeneous

ξC (x1, x3) = 0.2 × + 0.3 × + 0.4 × + 0.1 × = 0.213,
8 8 8 8 attribute information system. RH is a heterogeneous binary relation.
For any fuzzy decision-making object A ∈ F (U ) (where F (U ) stands for
2 1 1 1
ξC (x1, x 4 ) = 0.2 × + 0.3 × + 0.4 × + 0.1 × = 0.15. all fuzzy subsets of universe U) and x i ∈ U , the characteristic function of
8 8 8 8
the λ (0 ⩽ λ ≤ 1) heterogeneous relation classes with respect to the
This is the result of Definition 3.4. object set A is defined as follows:
Similarly, we have.
[xi]λ H 1, A (y ) ⩾ λ, y ∈ [x i]Rλ H ,
mC (x1, x1) = 0.2 × 1 + 0.3 × 1 + 0.4 × 1 + 0.1 × 1 = 1, χA R (x i ) = ⎧
⎨ 0, Others. (10)
⎩
mC (x1, x2) = 0.2 × 0 + 0.3 × 1 + 0.4 × 1 + 0.1 × 0 = 0.7,
In particular, if the decision-making object A ∈ P (U ) (where P (U )
mC (x1, x3) = 0.2 × 0 + 0.3 × 0 + 0.4 × 0 + 0.1 × 1 = 0.1, stands for all subsets of universe U), then the characteristic function of the

29
B. Sun et al. Computers & Industrial Engineering 122 (2018) 24–38

λ heterogeneous relation classes with respect to the object set A is For any crisp decision-making object A ∈ P (U ) and α (0 < α ≤ 1), the α
defined as follows: lower approximation Apr RαH (A) and the α upper approximation Apr αH (A)
R
with respect to (U , C , D , ) are defined as follows, respectively.
[xi]λ H 1, [x i]Rλ H ⊆ A,
χA R (x i ) = ⎧ l
⎨ 0, Others. ⎧ 1 [x ]λ H ⎫
⎩ (10′) Apr RαH (A) = x ∈ U ∑ χA R (x ) ⩾ α ,
⎨ l k=1
⎬
From the point of view of granular computing, the arbitrary binary ⎩ ⎭
relations over the universe of discourse are the key steps and the core l
⎧ 1 [x ]λ H ⎫
elements for establishing an expected rough set model. In this section, Apr αH (A) = x ∈ U
R ⎨ l
∑ (1−χ Ac R (x )) > 1−α
⎬
we successfully define the λ heterogeneous binary relation of multi- ⎩ k=1 ⎭
source heterogeneous attribute information system.
where Ac means the complementary set of A.
Here the parameter λ (0 ⩽ λ ≤ 1) is regarded as the acceptance
We then call the set-pair (Apr RαH (A), Apr RαH (A)) the α -multi-
threshold value of every binary preference relation related to concretely
decision-maker. Based on the above definitions, we know that there granulation rough fuzzy set if there is Apr RαH (A) ≠ Apr αH (A) for any
R
constructs a binary fuzzy relation over the universe of discourse with crisp decision-making object A ∈ P (U ) . Otherwise, A is called definable
different categories for the attribute. As we know, the similarity classes on (U , C , D , ) with respect to the λ− heterogeneous binary relation of
(or equivalence classes) is the key concept for constructing the basic the universe of discourse.
rough set model (Pawlak, 1982). Then, using the acceptance threshold As is well known, the original model of variable precision rough set was
value λ will obtain a heterogeneous similarity classes of any alternative defined by Ziarko (1993) and the precision parameter α was confined in
over multi-source heterogeneous attribute information system. So, we 0.5 < α ≤ 1. Because the prerequisite condition of 0.5 < α ≤ 1, then the
can define the multigranulation rough set based on the λ heterogeneous lower and upper approximation operators of Ziarko’s variable precision
binary relation classes. Subsequently, we will establish the hetero- rough set model keeping the good properties of the original Pawlak rough
geneous multigranulation rough set model over the multi-source in- set (Pawlak, 1982). Here, the precision parameter α used in the lower
formation system under the framework of multigranulation rough set approximation Apr RαH (A) and the upper approximation Apr αH (A) is con-
R
theory (Qian et al., 2010; Sun et al., 2018). fined to 0 < α ≤ 1 since the idea of the definition of the lower and upper
approximations comes from the background of risk decision-making under
uncertainty (Sun, Ma & Qian, 2017; Sun, Ma & Xiao, 2017). In order to
3.3. Rough approximation of a decision-making object over multi-source keep the consistency to the original concept and the convenience of the
heterogeneous attribute information system discussion in following sections as well as the good properties for the de-
fined model, we then confine parameter 0.5 < α ≤ 1. So, we call the set-
In this section, we will detailed discuss the rough approximation of a pair (Apr RαH (A), Apr αH (A)) variable precision multigranulation rough
R
decision-making object (a crisp and fuzzy decision-making object) over a fuzzy set based on λ− heterogeneous binary relation if there is
α
multi-source heterogeneous attribute information system (U , C , D , ) Apr R H (A) ≠ Apr RαH (A) for any object A ∈ P (U ) and 0.5 < α ≤ 1. Fur-
under the framework of multigranulation rough set (Qian et al., 2010; thermore, we present the positive region, negation region and the
Sun et al., 2018). As we know, there are two concretely opposite multi- boundary region of the variable precision multigranulation rough set model
granulation rough set models: optimistic and pessimistic multigranulation based on λ heterogeneous binary relation are as follows, respectively.
rough set models (Qian et al., 2010). In fact, the original two multi-
Pos RαH (A) = Apr RαH (A),
granulation rough set models are corresponding to risk preferring deci-
sion-making and risk-averse decision-making of reality (Sun, Ma & Qian, l
⎧ 1 [x ]λ H ⎫
2017; Sun, Ma & Xiao, 2017), respectively. However, it easy to see that Neg RαH (A) = U −Apr RαH (A) = x ∈ U ∑ (1−χ A R (x )) ⩾ α ,
the optimistic and pessimistic multigranulation rough set (Qian et al., ⎨ l k=1
⎬
⎩ ⎭
2010; Sun et al., 2018) only consider two extremely cases of a decision-
l
making process: completely risk-preferring and completely risk-averse. ⎧ 1 [x ]λ H ⎫
So, the way of the definition to construct the original multigranulation
Bn RαH (A) = x ∈ U
⎨
1−α <
l
∑ (1−χ Ac R (x )) < α .
⎬
⎩ k=1 ⎭
rough set have restricted the applications. Then, variable precision mul-
tigranulation rough set model (Sun et al., 2018) as a moderated im- In what follows, we deduce the optimistic and pessimistic multi-
proving version of the original models is proposed by introducing a pre- granulation rough fuzzy set models based on the Definition 3.9, respec-
cision parameter α (0 ⩽ α ≤ 1) . Meanwhile, both the optimistic and tively.
pessimistic multigranulation rough set model (Qian et al., 2010) can be
Remark 3.2. Let (U , C , D , ) be a multi-source heterogeneous attribute
derived by the variable precision multigranulation rough set model (Sun,
information system. For any crisp decision-making object A ∈ P (U )
Ma & Xiao, 2017), i.e., the original multigranulation rough set models are
and α (0.5 < α ≤ 1), Then
the special case of variable precision multigranulation rough set model.
Furthermore, the variable precision multigranulation rough set model 1
(1) If α = l , there are
could describe the decision-making process well than the original models.
Based on the above review of the relationship between the variable ⎧ 1
l [x ]λ H
1⎫
Apr RαH (A) = x ∈ U ∑ χA R (x ) ⩾ ,
precision multigranulation rough set and the original multigranulation ⎨ l l⎬
⎩ k=1 ⎭
rough set (Sun et al., 2018; Sun, Ma & Qian, 2017; Sun, Ma & Xiao,
l [x ]λ H
2017), this paper we only investigate the variable precision multi- ⎧ ⎫
= x∈U
⎨
∑ χA R (x ) ⩾ 1
⎬
granulation rough fuzzy set theory and methodology over multi-source ⎩ k=1 ⎭
heterogeneous attribute information system. Similar to the existing [x ]λ H
R
= {x ∈ U |χ A (x ) = 1, ∃ k = 1, 2, …, l}
results in Chen et al. (2012) and Slowinski and Vanderpooten (1990),
= {x ∈ U |[x ]λ H ⊆ A, ∃ k = 1, 2, …, l}
we also discuss the rough approximation of a crisp decision object and a R

fuzzy decision object over the multi-source heterogeneous attribute = {x ∈ U |[x ]1λΓ ⊆ A ∨ [x ]2λΓ ⊆ A ∨⋯∨ [x ]lλΓ ⊆ A, Γ = F , L, S }
R R R
information system, respectively. ≐ Apr ROH (A); (here the symbol ≐ meanings defined as).

Definition 3.9. Let (U , C , D , ) be a multi-source heterogeneous

attribute information system. RH is a heterogeneous binary relation.

30
B. Sun et al. Computers & Industrial Engineering 122 (2018) 24–38

⎧
l
[x ]λ H information system.
1 1⎫
Apr RαH (A) = x ∈ U
⎨ l
∑ (1−χ Ac R (x )) > 1− l
⎬
Remark 3.1 illustrates the interrelationship between variable pre-
⎩ k=1 ⎭ cision multigranulation rough fuzzy set and the optimistic/pessimistic
⎧ 1
l
[x ]λ H l − 1⎫
multigranulation rough fuzzy set over multi-source heterogeneous at-
= x∈U
⎨ l
∑ (1−χ Ac R (x )) > l ⎬ tribute information system. That is, both the optimistic and pessimistic
⎩ k=1 ⎭ multigranulation rough fuzzy set are the special case of variable pre-
l
⎧ [x ]λ H ⎫ cision multigranulation rough fuzzy set over multi-source hetero-
= x∈U
⎨
∑ (1−χ Ac R (x )) > l−1
⎬ geneous attribute information system. This conclusion is identical to the
⎩ k=1 ⎭ existing literatures (Qian et al., 2010; Sun et al., 2018; Sun, Ma & Qian,
l
⎧ [x ]λ H ⎫ 2017).
= x∈U l− ∑ χ Ac R (x ) > l−1
⎨ ⎬ In the following, we discuss the rough of a fuzzy decision-making
⎩ k=1 ⎭
object based on heterogeneous binary relation over multi-source het-
l
⎧ [x ]λ H ⎫ erogeneous attribute information system.
= x∈U
⎨
∑ χ Ac R (x ) < 1
⎬
⎩ k=1 ⎭ Definition 3.10. Let (U , C , D , ) be a multi-source heterogeneous
[x ]λ H
R
attribute information system. RH is a heterogeneous binary relation. For
= {x ∈ U |χ Ac (x ) = 0, ∀ k = 1, 2, …, l}
any fuzzy decision-making object A ∈ F (U ) and α (0 < α ≤ 1), the α
= {x ∈ U |[x ]kλΓ ¬ ⊆ Ac , ∀ k = 1, 2, …, l, Γ = F , L, S } lower approximation Apr RαH (A) and the α upper approximation
R
= {x ∈ U |[x ]kλΓ ¬ ⊆ U −A, ∀ k = 1, 2, …, l, Γ = F , L, S } Apr αH (A) with respect to (U , C , D , ) are defined as follows,
R R
respectively.
= {x ∈ U |[x ]1RλΓ ∩ A ≠ ∅ ∧ [x ]2λΓ ∩ A ≠ ∅ ∧ ⋯
R
l
1 [x ]λ H
∧ [x ]lλΓ ∩ A ≠ ∅, Γ = F , L, S } Apr RαH (A)(x ) = min{A (x )| ∑ χA R (x ) ⩾ α, x ∈ U },
R
l k=1
≐ Apr OH (A).
R
l
1 [x ]λ H
This is the heterogeneous binary relation-based optimistic multi- Apr RαH (A)(x ) = max{A (x )|
l
∑ χA R (x ) ⩾ α, x ∈ U }.
granulation rough set model. k=1

(2) If α = 1, there are We then call the set-pair (Apr RαH (A), Apr αH (A)) the
R
l
α -multigranulation fuzzy rough set if there is Apr αH (A) ≠ Apr αH (A)
[x ]λ H R R
⎧ 1 ⎫
Apr RαH (A) = x ∈ U l
∑ χA R (x ) ⩾ 1 , for any fuzzy decision-making object A ∈ F (U ) . Otherwise, A is called
⎨ ⎬
⎩ k=1 ⎭ definable on (U , C , D , ) with respect to the λ - heterogeneous binary
l
⎧ [x ]λ H ⎫ relation of the universe of discourse.
= x∈U
⎨
∑ χA R (x ) ⩾ l
⎬ It is easy to prove that the following conclusions are hold for the
⎩ k=1 ⎭
[x ]λ H
lower and upper approximations over multi-source heterogeneous at-
= {x ∈ U |χ A R (x ) = 1, ∀ k = 1, 2, …, l} tribute information system given in Definition 3.10.
= {x ∈ U |[x ]λ H ⊆ A, ∀ k = 1, 2, …, l}
R Theorem 3.1. Let (U , C , D , ) be a multi-source heterogeneous attribute
= {x ∈ U |[x ]1λΓ ⊆ A ∧ [x ]2λΓ ⊆ A ∧ ⋯ ∧ [x ]lλΓ ⊆ A, Γ = F , L, S } information system. RH is a heterogeneous binary relation. For any fuzzy
R R R
≐ Apr RPH (A); decision-making object A ∈ F (U ) and α (0 < α ≤ 1). Then

(1) Apr RαH (A) ⊆ Apr RαH (A) ⊆ U , (2) Apr RαH (∅) = Apr RαH (∅) = ∅,
l
[x ]λ H
(3) Apr RαH (Ac ) = (Apr αH (A))c , (4) Apr αH (Ac ) = (Apr RαH (A))c .
⎧ 1 ⎫ R R
Apr RαH (A) = x ∈ U
⎨ l
∑ (1−χ Ac R (x )) > 1−1 = 0
⎬ Similar to the Remark 3.2, we also can deduce the optimistic and pes-
⎩ k=1 ⎭ simistic multigranulation fuzzy rough set models based on the Definition
l
⎧ [x ]λ H ⎫ 3.10, respectively.
= x∈U
⎨
∑ (1−χ Ac R (x )) > 0
⎬
⎩ k=1 ⎭ Remark 3.3. Let (U , C , D , ) be a multi-source heterogeneous attribute
l
[x ]λ H
information system. For any fuzzy decision-making object A ∈ F (U )
⎧ ⎫
= x∈U l− ∑ χ Ac R (x ) > 0 and α (0 < α ≤ 1), Then
⎨ k=1
⎬
⎩ ⎭
1
⎧
l
[x ]λ H ⎫ (1) If α = l , there are
= x∈U
⎨
∑ χ Ac R (x ) < l
⎬ l
k=1 [x ]λ H
⎩ ⎭ Apr RαH (A)(x ) = min{A (x )| l
1 1
[x ]λ H
∑ χA R (x ) ⩾ l , x ∈ U },
R k=1
= {x ∈ U |χ Ac (x ) = 0, ∃ k = 1, 2, …, l}
l
[x ]λ H
= {x ∈ U |[x ]Rλ H ∩ A ≠ ∅, ∃ k = 1, 2, …, l} = min{A (x )| ∑ χ A R (x ) ⩾ 1, x ∈ U }
k=1
= {x ∈ U |[x ]1RλΓ ∩ A ≠ ∅∨ [x ]2RλΓ ∩ A ≠ ∅ ∨⋯ [x ]λ H
= min{A (x )|χ A R (x ) = 1, ∃ k = 1, 2, …, l, x ∈ U }
∨ [x ]lλ
RΓ
∩A ≠ ∅, Γ = F , L, S }
= min{A (x )|A (y ) ⩾ λ, y ∈ [x ]Rλ H , ∃ k = 1, 2, …, l, x , y ∈ U }
≐ P
Apr R H (A).
= {A (x )|Ay ∈ [x ]1λΓ (y ) ⩾ λ ∨ Ay ∈ [x ]2λΓ (y ) ⩾ λ ∨⋯∨ Ay ∈ [x ]lλΓ (y ) ⩾
R R R
This is the heterogeneous binary relation-based pessimistic multi-
λ, Γ = F , L, S;x , y ∈ U }
granulation rough set model.
≐ Apr ROH (A)(x );
Furthermore, we also can deduce other existing rough set models
Similarly, the upper approximation can be directly obtained as the
(Slowinski & Vanderpooten, 1990; Sun et al., 2018) based on hetero-
same way.
geneous binary relation over multi-source heterogeneous attribute

31
B. Sun et al. Computers & Industrial Engineering 122 (2018) 24–38

l 4. Heterogeneous multigranulation fuzzy rough set-based

1 [x ]λ H 1
Apr RαH (A)(x ) = min{A (x )| l ∑ χA R (x ) ⩾ l , x ∈ U }, approach to MAGDM
k=1
l
[x ]λ H
= max{A (x )| ∑ χ A R (x ) ⩾ 1, x ∈ U } In this section, based on the heterogeneous multigranulation fuzzy
k=1 rough established in Section 3.3, we will construct a new model and
= max{A (x )|Ay ∈ [x ]1λΓ (y ) ⩾ λ ∨ Ay ∈ [x ]2λΓ (y ) ⩾ approach to MAGDM with heterogeneous preference information. Also,
R R we will present the decision-making algorithm and the general steps for
λ ∨⋯∨ Ay ∈ [x ]lλΓ (y ) ⩾ λ, Γ = F , L, S;x , y the established method in detail.
R

∈ U}
4.1. The problem of MAGDM with heterogeneous preference information
≐ Apr OH (A)(x ).
R
Let U = {x1, x2 , …, x n} be a discrete set of alternatives,
This is the heterogeneous binary relation-based optimistic multi- C = {c1, c2, …, cm} be the set of attributes, ω = {ω1, ω2, …, ωm}T be the
granulation rough set model. weight vector of attributes where ωj ⩾ 0, ∑ j = 1 ωj = 1. Let
m

(2) If α = 1, there are D = {d1, d2, …, dl} be the set of decision makers which come from dif-
l ferent areas with different knowledge background and interesting, and
1 [x ]λ H
Apr RαH (A)(x ) = min{A (x )| l ∑ χA R (x ) ⩾ 1, x ∈ U }, μ = (μ1 , μ 2 , …, μl )T be the weight vector of decision-makers, where
k=1 l
μk ⩾ 0, k = 1, 2, …, l , and satisfies ∑i = 1 μk = 1. The decision-makers
l
[x ]λ H dk ∈ D present their preference evaluation for all alternatives about
= min{A (x )| ∑ χ A R (x ) ⩾ l, x ∈ U }
attribute set C with the interesting familiar way of themselves (i.e., the
k=1
[x ]λ H decision-makers can express their preference evaluation for every alternative
= min{A (x )|χ A R (x ) = 1, ∀ k = 1, 2, …, l, x ∈ U } by fuzzy values, linguistic values or symbol values, and etc. This paper
= min{A (x )|A (y ) ⩾ λ, y ∈ [x ]Rλ H , ∀ k = 1, 2, …, l, x , y ∈ U } confines three different attribute values given by decision-makers in order to
convenient the calculation).  = {fk (x i , cj )|i = 1, 2, …,
= min{A (x )|Ay ∈ [x ]1λΓ (y ) ⩾ λ ∧ Ay ∈ [x ]2λΓ (y ) ⩾
R R n; j = 1, 2, …, m; k = 1, 2, …, l} is a family of mapping sets between
λ ∧ ⋯ ∧ Ay ∈ [x ]lλΓ (y ) ⩾ λ, Γ = F , L, S } universe U and attribute set C, where fk : U × C → Vk (k = 1, 2, …, l) and
R Vk (k = 1, 2, …, l) is the domain of attribute set C with the available
≐ Apr RPH (A); information dk (dk ∈ D) . That is, ∀ x i ∈ U , cj ∈ C , dk ∈ D , fk (x i , cj ) ∈ Vk ,
i.e., fk (x i , cj ) is the evaluation of the alternative x i with respect to the
attribute cj according to the preference information dk such as fuzzy
values, linguistic values or symbol values. So, we construct a MAGDM
l
[x ]λ H
1 system with heterogeneous preference information (U , C , D ,  , ω, μ) .
Apr RαH (A)(x ) = max{A (x )| l ∑ χA R (x ) ⩾ 1, x ∈ U },
k=1 Then, the decision-making problem is how to determine the optimal
l
[x ]λ H
alternative or ranking all alternatives according to the evaluations
= max{A (x )| ∑ χ A R (x ) ⩾ l, x ∈ U } given by different decision-makers with different preference informa-
k=1 tion. This section will construct a new approach to deal with this kind of
= max{A (x )|Ay ∈ [x ]1λΓ (y ) ⩾ λ ∧ Ay ∈ [x ]2λΓ (y ) ⩾ decision-making problems by using the heterogeneous multigranulation
R R
fuzzy rough set established in Section 3.
λ ∧ ⋯ ∧ Ay ∈ [x ]lλΓ (y ) ⩾ λ, Γ = F , L, S }
R

≐ Apr RPH (A). 4.2. Decision-making method and process

This is the heterogeneous binary relation-based pessimistic multi- According to the characteristics of the MAGDM problem with het-
granulation rough set model. erogeneous preference information given in Section 4.1, we will present
the process of decision-making under the framework of heterogeneous
Similarly, we can deduce other new generalized rough set models multigranulation fuzzy rough set one by one.
based on Definitions 3.9 and 3.10 by selecting different combination of Firstly, constructing the fuzzy decision-making object of the uni-
the index of l and parameter α . Meanwhile, we also can establish several verse of all alternatives.
interesting properties and conclusions similar to Sun et al. (2014), Sun, Based on the basic principle of heterogeneous multigranulation
Ma and Xiao (2017), and Sun, Ma and Qian (2017). Furthermore, we also fuzzy rough set model, a fuzzy decision-making object is approximated
can find that the heterogeneous multigranulation rough fuzzy set is the over the multi-source heterogeneous attribute information system. As
special case of heterogeneous multigranulation fuzzy rough set according the assumption in Section 4.1, there are fuzzy values, linguistic values
to the Definitions 3.9 and 3.10. This conclusion is same to the original and symbol values for the attributes given by experts. The problem of
fuzzy rough set theory (Chen et al., 2012; Slowinski & Vanderpooten, decision-making over the multi-source heterogeneous attribute in-
1990). Therefore, we establish the theoretical basis of multigranulation formation system is ranking all alternatives according to the hetero-
fuzzy rough set over heterogeneous attribute decision information system. geneous preference information of all experts. So, we should define the
It is easy to find that the heterogeneous multigranulation fuzzy rough set fuzzy decision object under the conditional of the heterogeneous pre-
is more generalized than the existing multigranulation rough set models ference information of all experts. Then, the formulation of the mem-
(Qian et al., 2010; Sun et al., 2018; Sun, Ma & Qian, 2017; Sun, Ma & bership degree of any alternative with respect to a fuzzy decision-
Xiao, 2017) because the existing multigranulation rough set models deal making object over (U , C , D ,  , ω, μ) is as follows:
with single attribute values of the considered decision information m
system. Then, the proposed model has improved and enriched the original ⎧
⎪∑ ωt ct (x ), ct (x ) ∈ RF ;
multigranulation rough set theory (Qian et al., 2010; Sun et al., 2018) ⎪ t=1
both in theoretical and application aspects. Furthermore, it will provide a Ak (x ) = I (∑m ωt ct (x ) )
t=1
⎨ q
, ct (x ) ∈ RL ; ∀ x ∈ U , k = 1, 2, …, l;
new tool and idea to handle multi-source information system with in- ⎪ ∑m ω c (x )
accuracy, incomplete and uncertainty available data or other various ⎪ t =m1 t t , ct (x ) ∈ RS ;
decision information system with hybrid values of attribute set. ⎩ ∑t = 1 ct (x ) (11)

32
B. Sun et al.
Secondly, ranking all alternatives over MAGDM information sys-
tems with heterogeneous preference information (U , C , D ,  , ω, μ) .
As is well known, from the point of view of decision-making, the
lower and upper approximations of the original rough set corresponds
to the minimum and maximum risk decision-making principle of clas-
sical operational research (Huseyin & Warren, 2005). So, how to fusion
the information of the lower and upper approximations of a fuzzy de-
cision-making object with respect to MAGDM information with het-
erogeneous preference information (U , C , D ,  , ω, μ) and then obtain
the final optimal alternative are the key steps. From the Definition 3.10,
we know that the lower and upper approximations of a fuzzy decision-
making object over (U , C , D ,  , ω, μ) are also fuzzy set of universe U.
Thus, we use the fuzzy logical operators to fusion the information of the
lower and upper approximations. In the following, we introduce some
concepts of the fuzzy logical operators (Radzikowska & Kerre, 2002).
Triangular norms (or shortly t-norms) have been originally studied
within the framework of probabilistic metric spaces (Schweizer & Sklar,
1961, 1983). In this context, t-norms proved to be an appropriate tool
to deal with the information fusion of the lower and upper approx-
imations of any fuzzy decision-making object over (U , C , D ,  , ω, μ) .
Definition 4.1 (Radzikowska and Kerre, 2002). A t-norm is an
increasing, associative and commutative mapping
T: [0, 1] × [0, 1] → [0, 1] that satisfies the boundary condition
( ∀ x ∈ [0, 1], T (x , 1) = x ) . At the same time, a triangular conorm (or
shortly t-conorm) is an increasing, associative and commutative
mapping S: [0, 1] × [0, 1] → [0, 1] that satisfies the boundary
condition ( ∀ x ∈ [0, 1], S (x , 0) = x ) .In general, the most popular
continuous t-norms and t-conorms are as follows (Radzikowska &
Kerre, 2002), respectively.
(1) TM (x , y ) = min{x , y} (The standard min operator);
SM (x , y ) = max{x , y} (The standard max operator);
(2) TP (x , y ) = x ∗y (The algebraic product); SP (x , y ) = x + y−x ∗y (The
probabilistic sum);
(3) TL (x , y ) = max{0, x + y−1} (The Łukasiewicz t-norm);
SL (x , y ) = min{1, x + y} (The bounded sum).Based on the fuzzy
logical operators, we define the individual ranking function for any
alternative of the MAGDM with heterogeneous preference.
Definition 4.2. Let (U , C , D ,  , ω, μ) be the MAGDM system with
heterogeneous preference information. For the fuzzy decision-making
object Ak ∈ F (U )(k = 1, 2, …, l) determined by the preference
information of the kth expert and precision parameter α (0 < α ≤ 1),
we call
δk (x i ) = TL (Apr RαH (Ak )(x i ), Apr αH (Ak )(x i )), ∀ x i ∈ U ; k = 1, 2, …, l; i
R
= 1, 2, …, m.
the individual ranking function of expert k for the alternative x i with
the precision parameter α .
With the definition of individual ranking function of a hetero-
geneous preference-based MAGDM problem, the following conclusion
is clear.
Property 4.1.
0 ⩽ δk (x i )≤ 1, ∀ x i ∈ U , k = 1, 2, …, l; i = 1, 2, …, m.
Then, we can present the definition of the collective (group) ranking
function (or collective optimal index) for any alternative of the MAGDM
with heterogeneous preference.
Definition 4.3. Let (U , C , D ,  , ω, μ) be the MAGDM system with
heterogeneous preference information. For any fuzzy decision-making
object A ∈ F (U ) and α (0 < α ≤ 1), we call
l
δ (x i ) = ∑ μk δk (x i ), ∀ x i ∈ U ; i = 1, 2, …, m.
k=1
Computers & Industrial Engineering 122 (2018) 24–38
the collective optimal index for the alternative x i with the precision
parameter α .
Similarly, the following result is clear.
Property 4.2.
0 ⩽ δ (x i )≤ 1, ∀ x i ∈ U , i = 1, 2, …, m.
Thus, we can ranking all alternatives and then present the final
optimal decision-making for the MAGDM problem with heterogeneous
preference information by using the collective optimal index δ (x ) for
any x ∈ U . So far we establish an approach to MAGDM with hetero-
geneous preference information based on heterogeneous multi-
granulation fuzzy rough set.
In this paper, we investigate a kind of MAGDM problem with het-
erogeneous preference information by using the multigranulation rough
set theory (Sun et al., 2018). Generally speaking, the consensus
reaching process and the optimal alternative selection process are two
important issues in traditionally GDM problems. As is well known, the
existing approaches in the literatures (Wang, Xu, & Pedrycz, 2017; Wu,
Chiclana, Fujita, & Herrera-Viedma, 2017) use the measurement of si-
milarity or deviation measure and the aggregation operators to handle
of the individual consistency and collective consensus in the process of
GDM. However, one of the limitations is there could be obtained dif-
ferent optimal decision results when different aggregation operators
were used for the same multiple attribute group decision-making pro-
blems (MAGDM). In fact, the final optimal alternative (or decision-
making result) of a MAGDM problem could not based on the real pre-
ference evaluation of all experts since the necessary requirement of the
individual consistency and collective consensus. It can be easily seen
that this paper presents a different way to deal with the considered
GDM problem comparing to the existing literatures (Wang et al., 2017).
The remarkable difference is the method of dealing with the individual
preference and collective preference in the process of GDM. The pro-
posed heterogeneous multigranulation fuzzy set-based MAGDM ap-
proach explores a new perspective and pattern comparing to the ex-
isting literatures. The individual preference is fused by using the
multigranulation lower and upper approximations of a decision-making
object with respect to the approximation decision-making space. Then
the preference information given by every experts will be reserved as
much as possible in the process of GDM and then the final optimal al-
ternative will be the most suitable for real scenarios of the reality. Also,
the final optimal alternative (or decision-making result) is the lest un-
ique results for the considered problem of GDM.
4.3. Algorithm for the MAGDM with heterogeneous preference information
(12)
In this section, we present the steps of the proposed MAGDM model
with heterogeneous preference information based on heterogeneous
multigranulation fuzzy rough set theory as follows:
Input MAGDM system with heterogeneous preference information
(U , C , D ,  , ω, μ) ;
Output The sort ordering for all alternatives;
Step 1 Computing the λ−heterogeneous binary relation RH
according to formula (10);
Step 2 Constructing the fuzzy decision-making object
Ak (k = 1, 2, ⋯, l) according to formula (11);
Step 3 Computing the heterogeneous multigranulation lower
approximation Apr RαH (Ak )(x i ) and upper approximation
Apr αH (Ak )(x i ) , respectively;
R
Step 4 Computing the individual ranking function δk (x i ) using
formula (12);
(13)
33
B. Sun et al. Computers & Industrial Engineering 122 (2018) 24–38

Step 5 Computing the collective optimal index δ (x i ) using formula Table 3

(13); The heterogeneous attribute emergency plans evaluation GDM information
Step 6 Present the ranking according to the values of the collective system.
optimal index. U d1 d2

c1 c2 c3 c4 c5 c6 c7 c1 c2 c3 c4 c5 c6 c7
5. An illustrative example
x1 0.8 0.4 0.4 0.2 0.3 0.7 0.6 s2 s−3 s0 s1 s−2 s4 s−1
x2 0.3 0.7 0.7 0.2 0.5 0.4 0.3 s2 s3 s1 s0 s−4 s2 s−3
In this section, we apply the model and method of heterogeneous
x3 0.2 0.6 0.6 0.6 0.7 0.6 0.5 s−1 s0 s2 s2 s1 s−3 s1
multigranulation fuzzy rough set-based MAGDM to a problem of eva- x4 0.6 0.5 0.5 0.5 0.3 0.3 0.6 s3 s1 s−1 s0 s2 s−2 s2
luation of emergency plans for unconventional emergency events (Sun, x5 0.4 0.2 0.2 0.3 0.4 0.4 0.7 s−4 s−1 s2 s0 s1 s3 s1
Ma & Qian, 2017), which the problem of evaluation of emergency plans x6 0.2 0.5 0.5 0.5 0.8 0.5 0.9 s0 s2 s3 s1 s3 s−4 s4
is cited from Sun, Ma and Qian (2017) and makes some improvement
under the environment of MAGDM with heterogeneous preference in-
formation. Then, we will show the basic principle and steps of the Table 4
method established in this paper by discussing the illustrative example. (Continued Table 3) The heterogeneous attribute emergency plans evaluation
GDM information system.
The rapid growth in world population and increased human con-
centrations in dangerous environments have led to increases in both the U d3 d4
frequency and severity of unconventional emergency events (Sun & Ma,
c1 c2 c3 c4 c5 c6 c7 c1 c2 c3 c4 c5 c6 c7
2016). Then, emergency management is a new emerging study direc-
tion of management science in the recent years. As a result, emergency x1 1 1 2 2 3 1 3 0.4 0.2 0.5 0.4 0.4 0.7 0.9
decision-making becomes the key issue of emergency management of x2 2 3 1 2 3 2 1 0.7 0.5 0.4 0.7 0.3 0.5 0.8
unconventional emergency events and has been developed into a fra- x3 3 1 3 1 1 2 3 0.3 0.4 0.6 0.5 0.6 0.6 1
mework incorporating multiple disciplines, decision theories and x4 2 3 2 3 2 2 2 0.5 0.3 0.7 0.3 0.5 0.9 0.2
x5 1 2 3 2 3 3 1 0.4 0.4 0.9 0.6 0.2 1 0.4
methodologies (Sun & Ma, 2016; Xu, Zhang, & Wang, 2015). So far x6 2 3 2 1 2 3 2 0.8 0.7 0.1 0.7 0.1 0.8 0.6
many quantitative approaches have been used in studying the emer-
gency decision-making problems. One of these approaches is known as
plans-based decision-making (Jenkins, 2000). This idea depends on values, the preference evaluation of expert d3 are symbol values and the
emergency preparedness plans which are pre-established according to preference evaluation of expert d4 are fuzzy values, respectively. The
the characteristics of different unconventional emergency events (Sun & preference evaluation of all experts about all attributes for every
Ma, 2016). So, a new problem in the field of emergency decision- emergency plans are present in Tables 3 and 4, respectively. Mean-
making is how to evaluate pre-established emergency plans and then while, the weights of every expert are given as follows, respectively.
select the optimal one to handle concrete emergency events.
In general, several experts which come from different areas as well μ1 = 0.1, μ 2 = 0.3, μ3 = 0.4 μ4 = 0.2.
as different background knowledge and specialization will be invited to
evaluate a considered emergency preparedness in order to establish an Firstly, we calculate the λ−heterogeneous binary relation RH over
optimal alternative in reality. Then the experts will express the pre- universe U according to the principle and process given in Section 3.2.
ference evaluation or opinions by using their familiarize ways of their By Definition 3.1, we obtain the fuzzy equivalence relation R1F (C )
area of specialization, i.e., there will be multiple attribute values given and R4F (C ) over universe U according the preference evaluation given
by the invited experts. So, the problem of emergency plans evaluation by expert d1 and d4 as follows, respectively.
of unconventional emergency events is a multi-source heterogeneous
attribute GDM process. This section will present the model and method ⎛U x1 x2 x3 x4 x5 x6 ⎞
to deal with the evaluation of emergency plans by using the hetero- ⎜ x11 0.050 0.030 0.035 0.045 0.035 ⎟
geneous multigranulation fuzzy rough set-based MAGDM with hetero- ⎜ x2− 1 0.030 0.035 0.045 0.035 ⎟
geneous preference information. The statement of the problem is as R1F (C ) = ⎜ x3− − 1 0.045 0.035 0.045 ⎟ R4F (C )
follows. ⎜ x4− − − 1 0.040 0.050 ⎟
⎜ x5− − − − 1 0.040 ⎟
Let U = {x1, x2 , x3 , x 4 , x5 , x 6} be six emergency plans established in ⎜ ⎟
advance of an unconventional emergency event and C = ⎝ x6− − − − − 1 ⎠
{Comprehensiveness and completeness (c1), Feasibility (c2), Timely ⎛ U x1 x2 x3 x4 x5 x6 ⎞
Response (c3), Budgeting the cost (c4 ), Ability to adjust the plan (c5), ⎜ x1 1 0.035 0.045 0.030 0.040 0.035 ⎟
Effectiveness of handling (c6), Technical level of the equipment (c7)} be ⎜ x2 − 1 0.040 0.030 0.045 0.035 ⎟
the attribute set of the basic description of emergency plans. (For the = ⎜ x3 − − 1 0.020 0.040 0.040 ⎟
detailed interpretation and definition for the attribute set are suggested ⎜ x4 − − − 1 0.035 0.030 ⎟
to refer to Sun & Ma (2016).) The weights of every attribute are pre- ⎜ x5 − − − − 1 0.045 ⎟
⎜ ⎟
sented as follows, respectively. ⎝ x6 − − − − − 1 ⎠

ω1 = 0.2, ω2 = 0.1, ω3 = 0.2, ω4 = 0, 05, ω5 = 0.15, ω6 By Definitions 3.4 and 3.6, we can obtain the distance measure
= 0.2, ω7 = 0.1. ξC2 (x i , x j )(i, j = 1, 2, 3, 4, 5, 6) according to the linguistic value pre-
ference evaluation given by expert 2 and the similarity measure
Suppose there are four experts d1, d2, d3 and d4 which are invited to
mC3 (x i , x j )(i, j = 1, 2, 3, 4, 5, 6) according to the symbol value pre-
evaluate the six emergency plans according to the area of specialization
ference evaluation given by expert 3over universe U as follows, re-
of themselves. Here we suppose the preference evaluation of expert d1
spectively.
are fuzzy values, the preference evaluation of expert d2 are linguistic

34
B. Sun et al. Computers & Industrial Engineering 122 (2018) 24–38

ω1 c1 (x 1) + ω2 c 2 (x 1) + ω3 c3 (x 1) + ω4 c4 (x 1) + ω5 c6 (x 1) + ω7 c 7 (x 1)
⎛U x1 x2 x3 x4 x5 x6 ⎞ A3 (x1) = c1 (x 1) + c 2 (x 1) + c3 (x 1) + c4 (x 1) + c6 (x 1) + c 7 (x 1)
0 0.219 0.425 0.369 0.388 0.544 ⎟
⎜ x1 0.2 × 1 + 0.1 × 1 + 0.2 × 2 + 0.05 × 2 + 0.15 × 3 + 0.2 × 1 + 0.1 × 3 1.76
= = 0.1346.
−
⎜ x2 0 0.419 0.375 0.394 0.488 ⎟ 1+1+2+2+3+1+3 13
ξC2 (x i , x j ) = ⎜ x3
− − 0 0.256 0.280 0.181 ⎟ mC3 (x i , x j )
Then, we have
−
⎜ x4 − − 0 0.481 0.288 ⎟
⎜ x5
− − − − 0 0.419 ⎟ 0.1346 0.1393 0.1571 0.1344 0.15 0.15
⎜ ⎟ A3 = + + + + + .
−
⎝ x6 − − − − 0 ⎠ x1 x2 x3 x4 x5 x6
⎛ U x1 x2 x3 x 4 x5 x 6 ⎞ Similar to the expert d1, we can obtain the fuzzy decision-making
⎜ x1 1 0.2 0.2 0.2 0.4 0.2 ⎟ object determined by expert d4 as follows:
⎜ x2 − 1 0.2 0.5 0.3 0.3 ⎟
= ⎜ x3 − − 1 0.2 0.2 0.05⎟ 0.51 0.53 0.555 0.56 0.6 0.62
A4 = + + + + + .
⎜ x 4 − − − 1 0 0.75⎟ x1 x2 x3 x4 x5 x6
⎜ x5 − − − − 1 0.2 ⎟
⎜ ⎟ Suppose the precision parameter α = 0.45, then we can calculate the
⎝ x6 − − − − − 1 ⎠
lower and upper approximations of fuzzy decision-making object
Ak (k = ,1, 2, 3, 4) with respect to MAGDM system with heterogeneous
Suppose the threshold value parameter λ = 0.22 , then we can obtain
the λ−heterogeneous binary relation classes [x i]Rλ H (i = 1, 2, 3, 4, 5, 6) as preference information (U , C , D ,  , ω, μ) , respectively.
follows, respectively. 0.535 0.465 0.525 0.460 0.365
Apr R0.45
H (A1 ) = + + + +
x1 x2 x3 x4 x5
[x1]0.22
H = {[x1]1,0.22 = {x1}; [x1]2,0.22 = {x1, x2}; [x1]3,0.22 = {x1}; [x1]4,0.22 0.525 0.535 0.465 0.525
R RF RL RS RF
+ , Apr R0.45
H (A1 ) = + +
= {x1}; x6 x1 x2 x3
0.460 0.365 0.525
+ + + ,
[x2]0.22 = {[x2]1,0.22 = {x2}; [x2]2,0.22 = {x1, x2}; [x2]3,0.22 = {x2}; [x2]4,0.22 x4 x5 x6
RH RF RL S RF R

= {x2}; 0.2375 0.2375 0 0 0 0

Apr R0.45
H (A2 ) = + + + + + , Apr R0.45
H (A2 )
x1 x2 x3 x4 x5 x6
[x3]0.22 = {[x3]1,0.22 = {x3}; [x3]2,0.22 = {x3 , x 6}; [x3]3,0.22 = {x3}; [x3]4,0.22 0.2375 0.2375 0 0 0 0
RH RF RL S RF R = + + + + + ,
x1 x2 x3 x4 x5 x6
= {x3};
0 0 0 0 0 0
Apr R0.45
H (A3 ) = + + + + + , Apr R0.45
H (A3 )
[x 4]0.22 = {[x 4]1,0.22 = {x 4}; [x 4]2,0.22 = {x 4}; [x 4]3,0.22 = {x 4}; [x 4]4,0.22 x1 x2 x3 x4 x5 x6
RH RF RL S RF R
0 0 0 0 0 0
= {x 4}; = + + + + + ,
x1 x2 x3 x4 x5 x6

[x5]0.22 = {[x5]1,0.22 = {x5}; [x5]2,0.22 = {x5}; [x5]3,0.22 = {x5}; [x5]4,0.22 0.51 0.53 0.555 0.56 0.66
RH RF RL S RF Apr R0.45
H (A 4 ) = + + + +
R
x1 x2 x3 x4 x5
= {x5};
0.52 0.45 0.51 0.53 0.555 0
+ , Apr H (A 4 ) = + + +
x6 R x1 x2 x3 x4
[x 6]0.22
RH
= {[x 6]1,0.22
RF
= {x 6}; [x 6]2,0.22
RL
= {x3 , x 6}; [x 6]3,0.22
S = {x 6}; [x 6]4,0.22
RF R 0 0.62
+ + .
= {x 6}. x5 x6

Subsequently, we calculate the fuzzy decision-making object So, we can obtain the individual ranking function
Ak (k = 1, 2, 3, 4) determined by the preference information of every δk (x i )(k = 1, 2, 3, 4; i = 1, 2, 3, 4, 5, 6) using the Definition 4.2 as fol-
expert over (U , C , D ,  , ω, μ) by formula (11), respectively. lows, respectively.
For the expert d1, we have
δ1 (x1) = TL (Apr R0.45 0.45 0.45
H (A1 )(x1), Apr H (A1 )(x1)) = max{0, Apr R H (A1 )(x1)
R
A1 (x1) = ω1 c1 (x1) + ω2 c2 (x1) + ω3 c3 (x1) + ω4 c4 (x1) + ω5 c6 (x1) + Apr 0.45
H (A1 )(x1)−1} = 0.05.
R
+ ω7 c7 (x1),
Then, we have
= 0.2 × 0.8 + 0.1 × 0.4 + 0.2 × 0.4 + 0.05 × 0.2 + 0.15 × 0.3
+ 0.2 × 0.7 + 0.1 × 0.6 = 0.535; δ1 (x1) = 0.07, δ1 (x2) = 0, δ1 (x3) = 0.05, δ1 (x 4 ) = 0, δ1 (x5)

So, we have = 0, δ1 (x 6) = 0.05;

0.535 0.465 0.525 0.46 0.365 0.525 δ2 (x1) = 0, δ2 (x2) = 0, δ2 (x3) = 0, δ2 (x 4 ) = 0, δ2 (x5)
A1 = + + + + + .
x1 x2 x3 x4 x5 x6 = 0, δ2 (x 6) = 0;
For the expert d2 , we have
δ3 (x1) = 0, δ3 (x2) = 0, δ3 (x3) = 0, δ3 (x 4 ) = 0, δ3 (x5)
I (ω1 c1 (x 1) + ω2 c 2 (x 1) + ω3 c3 (x 1) + ω4 c4 (x 1) + ω5 c6 (x 1) + ω7 c 7 (x 1))
A2 (x1) = q
= 0, δ3 (x 6) = 0;
I (0.2 × s2 + 0.1 × s−3 + 0.2 × s0 + 0.05 × s1 + 0.15 × s−2 + 0.2 × s4 + 0.1 × s−1)
= 4 δ4 (x1) = 0.02, δ4 (x2) = 0.06, δ4 (x3) = 0.11, δ4 (x 4 ) = 0.12, δ4 (x5)
I (s0.55) 0.55
= 4
= 4
= 0.2375; = 0.2, δ4 (x 6) = 0.24;

Then, we have Finally, we obtain the collective optimal index

δ (x i )(i = 1, 2, 3, 4, 5, 6) based on Definition 4.3 as follows.
0.2375 0.2375 0.2125 0.15 0.0875 0.2125
A2 = + + + + + . δ (x1) = μ1 δ1 (x1) + μ 2 δ2 (x1) + μ3 δ3 (x1) + μ4 δ4 (x1) = 0.011;
x1 x2 x3 x4 x5 x6

For the expert d3 , we have δ (x2) = μ1 δ1 (x2) + μ 2 δ2 (x2) + μ3 δ3 (x2) + μ4 δ4 (x2) = 0.012;

35
B. Sun et al.
δ (x3) = μ1 δ1 (x3) + μ 2 δ2 (x3) + μ3 δ3 (x3) + μ4 δ4 (x3) = 0.027;
δ (x 4 ) = μ1 δ1 (x3) + μ 2 δ2 (x 4 ) + μ3 δ3 (x 4 ) + μ4 δ4 (x 4 ) = 0.024;
δ (x5) = μ1 δ1 (x 4 ) + μ 2 δ2 (x5) + μ3 δ3 (x5) + μ4 δ4 (x5) = 0.04;
δ (x 6) = μ1 δ1 (x5) + μ 2 δ2 (x 6) + μ3 δ3 (x 6) + μ4 δ4 (x 6) = 0.053.
Based on the values of the collective optimal index
δ (x i )(i = 1, 2, 3, 4, 5, 6) , we then obtain the sorting for all emergency
plans as follows:
δ (x 6) ≻ δ (x5) ≻ δ (x3) ≻ δ (x 4 ) ≻ δ (x2) ≻ δ (x1).
Therefore, we complete the decision-making process for the pro-
blem of evaluation of emergency plans for unconventional emergency
events under the framework of heterogeneous multigranulation fuzzy
rough set theory.
Furthermore, we make some numerical analysis to justify the pro-
posal well.
(I) Taking λ = 0.22 and α = 0.80 . Then, we
δ (x 6) ≻ δ (x5) ≻ δ (x3) ≻ δ (x 4 ) ≻ δ (x2) ≻ δ (x1) .
(II) Taking λ = 0.4 and α = 0.45. Then, we
δ (x 6) ≻ δ (x5) ≻ δ (x3) ≻ δ (x 4 ) ≻ δ (x2) ≻ δ (x1) .
From the above cases discussion, we known that the results given by
the proposed model and method in Section 4 is more robust. So, the
established model is validity and effective. In fact, with the constantly
enriched means of information storage, representation and calculation
under the environment of big data, heterogeneous preference in-
formation group decision making problems are emerged in many areas
of industrial engineering. So, the proposed model and method in this
paper can be widely used to solve the problem of optimizing decision-
making in many industrial fields such as the bidding for complex con-
struction projects, the risk financial decision-making and the human
resource management.
Comparing to the traditional approach to MAGDM problems, the
heterogeneous attribute-based approach proposed in this paper has
several distinguishing characteristics. First, it provides another way to
acquire the collective preference based on the individual preference for
a considered GDM problem by using the lower and upper approxima-
tions of the heterogeneous attribute multigranulation rough set. In the
proposed model, the individual preference given by different decision-
maker is regarded as a granularity over the multigranulation approx-
imation decision-making space, and then the collective preference is
corresponding to the total granularity which is the fusion of all single
granularity (i.e., corresponds to individual preference) according to the
principle of the multigranulation lower and upper approximation. The
decision-making process given in this paper do not use the operator
aggregation but use the idea of the mutligrannlation rough set. That is,
the multigranulation rough set-based MAGDM model and method is a
non-operator aggregation process. Then the collective preference is
obtained by using the methodology of multigranulation rough set but
not the operator aggregation. As is well known, there is only a pair of
lower and upper approximations for any multigranulation rough set
model. So, the total granularity (i.e., the collective preference) is cer-
tain whenever any multigranulation rough set model is used. Therefore,
this way can avoid the problem that different optimal decision-making
results will be given when different aggregation operators are used for
the same group decision-making problem (Wang et al., 2017). Second,
heterogeneous structure is build based on different categories attribute
generated the binary relation classes and then it are regarded as the
preliminary granular during the process of construct the heterogeneous
multigranulation lower and upper approximations of any decision-
making objects over the universe of discourse. This process reserves the
original preference information of every decision-maker as much as
possible under the framework of multigranulaition rough set theory
Computers & Industrial Engineering 122 (2018) 24–38
when computing the final optimal result. So, the final solution will be
suitable for the scenario of the decision-making problem as much as
possible. At the same time, Chen, Zhang, and Dong (2015) presents a
survey regarding heterogeneous GDM and points out that there are
three types of fusion approaches for the heterogeneous preference in-
formation in GDM: the indirect approach, the optimization-based ap-
proach and the direct approach. Then, the model and method given in
this paper can be regarded a new way of the direct approach to het-
erogeneous GDM using the heterogeneous multigranulation fuzzy
rough set. Third, the decision-making object is regarded as a fuzzy
subset of the universe of discourse related to all alternatives of the
universe, the outranking sorting for all alternatives by combining risk
decision-making principle of classical operational research and the
heterogeneous attribute multigranulation rough fuzzy set. This could
make the final ranking for all alternatives reducing the subjectively
affects as possible as. Furthermore, this will substantially improve the
effectiveness of the decision-making process and the successful im-
plementation of the final solution.
have the 6. Conclusions
have the Decision-making, as a scientific theory and approach to deal with
the problems of management science in reality having a thousand-year
history, it can scientifically instruct and assist people to make reason-
able and effective ways to cope with the complexity scenarios of
practice. GDM methodology, as one of important theoretical tools,
which several experts provide their judgement or opinions over a set of
candidate alternatives, is used to handle the decision problems in the
increasing complexity of the socio-economic environment. So far a
large number of theoretical models and approaches have established to
solve various problems of GDM (Chauhan & Singh, 2016; Hwang & Lin,
1987; Inan et al., 2017; Lima-Junior & Carpinetti, 2016; Maio et al.,
2016; Shen et al., 2016). Generally speaking, traditional models and
approaches to GDM problems mainly include two processes: the con-
sensus process and selection process. The former refers to how to obtain
the maximum degree of consensus or agreement among the experts on
the solution alternatives. The latter consists of how to obtain the so-
lution set of alternatives from the opinions about the alternatives given
by the experts. So, it is preferable that the experts achieve agreement
among their preferences before applying the selection process (Sun, Ma
& Qian, 2017). It is easy to know that both the consensus process and
selection process are highly depended on the used aggregation opera-
tors while the preference or opinions of experts were aggregated.
Though many large-scale complexities and unstructured problems of
decision-making in management science of reality have solved based on
the existing models and approaches of GDM (Dong, Ding, Martínez, &
Herrera, 2017; Franco & Lord, 2011; Li, Dong, Herrera, Herrera-
Viedma, & Martínez, 2017; Wu et al., 2017), there could be obtained
different optimal decision results when different aggregation operators
were used for the GDM problems. So, a meaningful and valuable topic is
exploring another theory and then establishing new models and ap-
proaches to deal with GDM problems.
Rough set theory provides a new theory and tool to handle the
decision-making problems under uncertainty. This paper presents a
theory and method to deal with MAGDM problem with heterogeneous
preference information by using the heterogeneous multigranulation
fuzzy rough set theory. Under the framework of multigranulation rough
set theory, the multi-source heterogeneous preference information
provided by all experts are fused by the multigranulation lower and
upper approximations of any decision object. then, the final optimal
conclusion is given for the considered MAGDM problem with hetero-
geneous preference information. Furthermore, the proposed model and
method is applied to the problem of emergency management decision-
making of unconventional emergency events. The proposed model and
method in this paper is an initial attempt to introduce the generalized
rough set theory into the area of GDM. On the one hand, the generalized
36
B. Sun et al.
rough set-based GDM theory provides a new perspective and opinions
to the existing literatures. On the other hand, the model and approach
are a meaningful supplement and perfection of the original theories.
It could not be an absolute optimal decision-making for a manage-
ment science problems in reality, as well as for the emergency decision-
making of unconventional emergency events. Meanwhile, due to the
difference in theoretical basis and processing methods, there still lacks
an evaluation criterion strictly to assess decision results based on var-
ious theoretical tools. As is well known, the emergency decision-making
of unconventional emergency events is confronted with insufficient
information and limited time and the height of a psychological pressure
for decision-makers. This paper builds a new approach to emergency
decision-making problems of unconventional emergency events based
on heterogeneous multigranulation fuzzy rough set theory. The studies
of this paper are focusing on the basis of the theoretical aspect and the
general framework of decision-making process of the proposed model
and method. Therefore, it is recommended that the further improved of
the proposed method to apply more complexity decision-making pro-
blems as mentioned in above possible areas and the real-life data be
used to test the approach established in this paper. At the same time,
this paper only considers the complete heterogeneous preference in-
formation of a MAGDM problems. In fact, there are various incomplete
preference and hesitant information incorporated in a given MAGDM
(Xu, Li, & Wang, 2014; Xu, Chen, Rodríguez, Herrera, & Wang, 2016).
Therefore, further studies will consider incomplete heterogeneous
preference information or hesitant heterogeneous preference informa-
tion based on heterogeneous multigranulation rough set theory.
Acknowledgements
The authors are very grateful to the Editor-in-Chief Professor
Mohamed Dessouky, and the three anonymous referees for their
thoughtful comments and valuable suggestions. Some remarks directly
benefit from the referees’ comments. The work was partly supported by
the National Natural Science Foundation of China (71571090,
61772019), the National Science Foundation of Shaanxi Province of
China (2017JM7022), the Key Strategic Project of Fundamental
Research Funds for the Central Universities (JBZ170601), the
Interdisciplinary Foundation of Humanities and Information
(JB180608).
References
Abastante, F., Bottero, M., Greco, S., & Lami, I. (2014). Addressing the location of un-
desirable facilities through the dominance-based rough set approach. Journal of Multi-
Criteria Decision Analysis, 21(1-2), 3–23.
Chakhar, S., Ishizaka, A., Labib, A., & Saad, I. (2016). Dominance-based rough set ap-
proach for group decision. European Journal of Operational Research, 251, 206–224.
Chauhan, A., & Singh, A. (2016). A hybrid multi-criteria decision making method ap-
proach for selecting a sustainable location of healthcare waste disposal facility.
Journal of Cleaner Production, 139, 1001–1010.
Chen, X., Zhang, H., & Dong, Y. (2015). The fusion process with heterogeneous preference
structures in group decision making: A survey. Information Fusion, 24, 72–83.
Chen, D., Zhang, L., Zhao, S., Hu, Q., & Zhu, P. (2012). A novel algorithm for finding
reducts with fuzzy rough sets. IEEE Transactions on Fuzzy Systems, 20(2), 385–389.
Dong, Y., Ding, Z., Martínez, L., & Herrera, F. (2017). Managing consensus based on
leadership in opinion dynamics. Information Sciences, 397–398, 187–205.
Dong, Y., Li, C., & Herrera, F. (2016). Connecting the linguistic hierarchy and the nu-
merical scale for the 2-tuple linguistic model and its use to deal with hesitant un-
balanced linguistic information. Information Sciences, 367–368, 259–278.
Dong, Y., Zhang, H., & Herrera-Viedma, E. (2016). Consensus reaching model in the
complex and dynamic MAGDM problem. Knowledge Based Systems, 106, 206–219.
Dubois, D., & Prade, H. (1990). Rough sets and fuzzy rough sets. International Journal of
General Systems, 17, 191–209.
Franco, L., & Lord, E. (2011). Understanding multi-methodology: evaluating the per-
ceived impact of mixing methods for group budgetary decisions. Omega, 39, 362–373.
Huseyin, T., & Warren, B. (2005). A distributed decision-making structure for dynamic
resource allocation using nonlinear functional approximations. Operational Research,
53(2), 281–297.
Hwang, C. L., & Lin, M-J. (1987). Group decision making under multiple criteria: Methods and
applications. Berlin, New York: Springer-Verlag.
Inan, U., Gul, S., & lmaz, H. Y. (2017). A multiple attribute decision model to compare the
37
Computers & Industrial Engineering 122 (2018) 24–38
firms occupational health and safety management perspectives. Safety Science, 91,
221–231.
Ishizaka, A., & Nemery, P. (2013). A multi-criteria group decision framework for partner
grouping when sharing facilities. Group Decision and Negotiation, 22(4), 773–799.
Ishizaka, A., & Nemery, P. (2013). Multi-criteria decision analysis: methods and software.
Wiley.
Jenkins, I. (2000). Selecting scenarios for environmental disaster planning. European
Journal of Operational Research, 121(2), 275–286.
Kong, Y., Zhang, M., & Ye, D. (2016). A belief propagation-based method for task allo-
cation in open and dynamic cloud environments. Knowledge Based Systems, 115,
123–132.
Kucukaltan, B., Irani, Z., & Aktas, E. (2016). A decision support model for identification
and prioritization of key performance indicators in the logistics industry. Computers in
Human Behavior, 65, 346–358.
Li, C., Dong, Y., Herrera, F., Herrera-Viedma, E., & Martínez, L. (2017). Personalized
individual semantics in computing with words for supporting linguistic group deci-
sion making: An application on consensus reaching. Information Fusion, 33, 29–40.
Lima-Junior, F., & Carpinetti, L. (2016). A multicriteria approach based on fuzzy QFD for
choosing criteria for supplier selection. Computers in Industrial Engineering, 101,
269–285.
Maio, C., Fenza, G., & Loia, V. (2016). A framework for context-aware heterogeneous
group decision making in business processes. Knowledge Based Systems, 102, 39–50.
Ma, X., Zhan, J., Ali, M. I., & Mehmood, N. (2018). A survey of decision making methods
based on two classes of hybrid soft set models. Artificial Intelligence Review, 49(4),
511–529.
Onar, S., Oztaysi, B., Otay, I., & Kahraman, C. (2015). Multi-expert wind energy tech-
nology selection using interval-valued intuitionistic fuzzy sets. Energy, 90(Part 1),
274–285.
Özgür, K., & Bilal, E. (2017). Multiple attribute group decision making: A generic con-
ceptual framework and a classification scheme. Knowledge Based Systems, 123, 13–30.
Pawlak, Z. (1982). Rough sets. International Journal of Information Sciences, 11(5),
341–356.
Qian, Y., Liang, J., Yao, Y., & Dang, C. (2010). A multigranulation rough set. Information
Sciences, 180, 949–970.
Radzikowska, A., & Kerre, E. (2002). A comparative study of fuzzy rough sets. Fuzzy Sets
and Systems, 126, 137–155.
Savage, L. (1954). The foundations of statistics. New York: Wiley.
Schweizer, B., & Sklar, A. (1961). Associative functions and statistical triangle inequal-
ities. Publications Mathematicae-Debrecen, 8, 169–186.
Schweizer, B., & Sklar, A. (1983). Probabilistic metric spaces. Amsterdam: North-Holland.
Shen, F., Xu, J., & Xu, Z. (2016). An outranking sorting method for multi-criteria group
decision making using intuitionistic fuzzy sets. Information Sciences, 334–335,
338–353.
Slowinski, R., & Vanderpooten, D. (1990). A generalized definition of rough approx-
imations based on similarity. IEEE Transactions on Knowledge and Data Engineering,
12(2), 331–336.
Sun, B., & Ma, W. (2016). An approach to evaluation of emergency plans for un-
conventional emergency events based on soft fuzzy rough set. Kybernetes, 45(3),
1–26.
Sun, B., Ma, W., & Chen, D. (2014). Rough approximation of a fuzzy concept on a hybrid
attribute information system and its uncertainty measure. Information Sciences, 284,
60–80.
Sun, B., Ma, W., Li, B., & Li, X. (2018). Three-way decisions approach to multiple attribute
group decision making with linguistic information-based decision-theoretic rough
fuzzy set. International Journal of Approximate Reasoning, 93, 424–442.
Sun, B., Ma, W., & Qian, Y. (2017). Multigranulation fuzzy rough set over two universes
and its application to decision making. Knowledge Based Systems, 123, 61–74.
Sun, B., Ma, W., & Xiao, X. (2017). Three-way group decision making based on multi-
granulation fuzzy decision-theoretic rough set over two universes. International
Journal of Approximate Reasoning, 81, 87–102.
Sun, B., Ma, W., & Zhao, H. (2016). Rough set-based conflict analysis model and method
over two universes. Information Sciences, 372, 111–125.
Ünlüç, H., Ç Ervural, B., Ervural, B., & Kabak, Ö. (2017). Cumulative belief degrees ap-
proach for assessment of sustainable development. In C. Kahraman, & U. Sari (Eds.).
Intelligence Systems in Environmental Management: Theory and Applications (pp. 257–
289). Cham: Springer International Publishing.
Wang, C., He, Q., Shao, M., & Hu, Q. (2017). Feature selection based on maximal
neighborhood discernibility. International Journal of Machine Learning Cybernetics.
http://dx.doi.org/10.1007/s13042-017-0712-6.
Wang, C., Hu, Q., Wang, X., Chen, D., & Qian, Y. (2017). Feature selection based on
neighborhood discrimination index. IEEE Transactions on Neural Networks and
Learning Systems, 00. http://dx.doi.org/10.1109/TNNLS.2017.2710422.
Wang, C., Shao, M., He, Q., Qian, Y., & Qi, Y. (2016). Feature subset selection based on
fuzzy neighborhood rough sets. Knowledge Based Systems, 111(1), 173–179.
Wang, H., Xu, Z., & Pedrycz, W. (2017). An overview on the roles of fuzzy set techniques
in big data processing: Trends, challenges and opportunities. Knowledge Based
Systems, 118(15), 15–30.
Wu, J., Chiclana, F., Fujita, H., & Herrera-Viedma, E. (2017). A visual interaction con-
sensus model for social network group decision making with trust propagation.
Knowledge Based Systems, 122, 39–50.
Wu, W., & Leung, Y. (2011). Theory and applications of granular labelled partitions in
multi-scale decision tables. Information Sciences, 181, 3878–3897.
Xu, Y., Cabrerizo, F., & Herrera-Viedma, E. (2017). A consensus model for hesitant fuzzy
preference relations and itsapplication in water allocation management. Applied Soft
Computing, 58, 265–284.
Xu, Y., Chen, L., Rodríguez, R., Herrera, F., & Wang, H. (2016). Deriving the priority
B. Sun et al.
weights from incomplete hesitant fuzzy preference relations in group decision
making. Knowledge Based Systems, 99, 71–78.
Xu, Y., Li, K., & Wang, H. (2013). Distance-based consensus models for fuzzy and mul-
tiplicative preference relations. Information Sciences, 253(20), 56–73.
Xu, Y., Li, K., & Wang, H. (2014). Incomplete interval fuzzy preference relations and their
applications. Comput. Ind. Eng. 67, 93–103.
Xu, Y., Zhang, W., & Wang, H. (2015). A conflict-eliminating approach for emergency
group decision of unconventional incidents. Knowledge Based Systems, 83, 92–104.
Yan, J. (1998). Theory of measures. Beijing: Science Press.
Yao, J., Vasilakos, A., & Pedrycz, W. (2013). Granular computing: perspectives and
challenges. IEEE Transactions on Cybernetics, 43(6), 1977–1989.
Yao, Y., Wong, S., & Wang, L. (1995). A non-numeric approach to uncertain reasoning.
38
Computers & Industrial Engineering 122 (2018) 24–38
International Journal of General Systems, 23, 343–359.
Zadeh, L. (1971). Similarity relations and fuzzy ordering. Information Sciences, 3,
177–200.
Zadeh, L. A. (1997). Toward a theory of fuzzy information granulation and its centrality
in human reasoning and fuzzy logic. Fuzzy Sets and Systems, 19(2), 111–127.
Zhan, J., & Alcantud, J. C. R. (2018). A novel type of soft rough covering and its appli-
cation to multicriteria group decision making. Artificial Intelligence Review. http://dx.
doi.org/10.1007/s10462-018-9617-3.
Zhou, L., & Wu, W. (2008). On generalized intuitionistic fuzzy approximation operators.
Information Sciences, 178, 2448–2465.
Ziarko, W. (1993). Variable precision rough sets model. Journal of Computer and Systems
Sciences, 46(1), 39–59.