Professional Documents
Culture Documents
1 s2.0 S0957417422017213 Main
1 s2.0 S0957417422017213 Main
Keywords: In the field of decision making, three-way decision making has been proven more fruitful for providing
Three-way multi-attribute decision making scopes to make delayed decision along with acceptance and rejection simultaneously. As a result, the decision
Incomplete information system risk and loss, which could occur due to take rapid decisions in traditional two-way decision making, are
Interval-valued Fermatean connection number
effectively reduced. Accordingly, this paper offers a novel three-way multi-attribute decision making model by
Possibility dominance relation
combining three-way decision making and multi-attribute decision making under an incomplete information
Relative utility function
Regret theory
system. The incertitude in the information system is illustrated by introducing interval-valued Fermatean
connection number based on interval-valued Fermatean fuzzy number and set pair analysis theory. Thereafter,
the achievement of this study is five-fold. First, a possibility dominance relation is developed under incomplete
information system in the basis of possibility degree measure of interval-valued Fermatean connection number.
Second, we put forward a novel procedure to set up the fuzzy state set. Third, the conditional probability is
estimated in light of the fuzzy state set and probability dominance relation. Fourth, the relative utility functions
are determined in the virtue of regret theory. Lastly, a three-way multi-attribute decision making model is
implemented in incomplete information system and exploited to deal with incomplete multi-attribute decision
making problems. Eventually, the propriety, stability and superiority of the proposed model is established via
conducting the comparative and experimental analysis.
1. Introduction maximum utility, (ii) 3WMADM models differentiate the objects into
three domains. However, most of the existing 3WMADM models were
In contrast to typical two-way decision making, three-way decision demonstrated in complete environments (Jia & Liu, 2021; Wang, Li,
making (3WDM) provides a new decision scope, namely delayed de- Qian et al., 2020; Wang et al., 2022a), and the 3WMADM models which
cision alongside acceptance and rejection, which effectively cut down were set up in incomplete environments (Xin et al., 2021; Yang, Li
the risk of decision. Consequently, since three-way decision was put et al., 2020; Zhan et al., 2021), cannot effectively deal with real-world
forth by Yao (2010) with the aid of the concept of decision-theoretic complex and modern situations. For example, the case studies given in
rough set (Pawlak, 1982), it has gained expansive attention in various Section 6, cannot be solved by the above-mentioned models. Therefore,
practical areas such as medical diagnosis (Li et al., 2020), investment there is an immediate need to build an unprecedented 3WMADM
evaluation (Jia & Liu, 2019), target threat assessment (Gao et al., 2020), model to deal with incomplete MADM (IMADM) problems that can
conflict analysis (Li, Wang et al., 2021), face recognition (Li et al., efficaciously resolve the modern decision making issues.
2016), and so forth. On the other hand, multi-attribute decision making Ordinarily, the information in an IMADM problem are published by
(MADM) is very serviceable in order to sort the alternatives on the
an incomplete information system (IIS). In real-world decision making
basis of its competency with respect to multiple attributes. Thus, the
problems, the information regarding the features of objects may not
three-way MADM (3WMADM), which is an amalgamation of 3WDM
always possible to record and those are often partial or not specified.
and MADM, has become a hot topic of the research ground because
Such type of information system is said to be an IIS. There are mainly
of the following advantages: (i) 3WMADM models encounter the losses
three approaches to tackle IIS (Du & Hu, 2016): (i) by filling the missing
or utilities incurred for taking different actions under various states,
values (Xin et al., 2021; Xu et al., 2020; Ye et al., 2021), (ii) by deleting
and the final decision is conducted based on the minimum loss or
∗ Corresponding author.
E-mail addresses: arijitmondal94math@gmail.com (A. Mondal), roysank@mail.vidyasagar.ac.in (S.K. Roy), dpamucar@gmail.com (D. Pamucar).
1
All authors are equally contributed to the manuscript in all sections.
https://doi.org/10.1016/j.eswa.2022.118688
Received 27 February 2022; Received in revised form 28 July 2022; Accepted 22 August 2022
Available online 28 August 2022
0957-4174/© 2022 Elsevier Ltd. All rights reserved.
A. Mondal et al. Expert Systems With Applications 211 (2023) 118688
or ignoring the missing values, and (iii) by treating the missing values IVFFNs may produce the information loss and inconsistent outcomes
as special symbols and do not concern about their earmarked values. (see Section 7.4). Motivated by these facts, in this study, we introduce
Because of the limitation of the first two approaches in inevitably a new concept of interval-valued Fermatean CN (IVF-CN) of the form
destroying the originality of data, we aim to implement the third 𝑎+𝑏𝑖+𝑐𝑗 in light of IVFFN and SPA theory, which satisfies the condition
approach in this study. In general, there are two types of missing values 𝑎3 + 𝑏3 + 𝑐 3 = 1. The main advantages of the proposed IVF-CN are:
presented in an IIS: the lost values and the absent values. For simplicity, (i) it combines certainty and uncertainty in a consolidated system, and
we mainly focus on envisaging the lost values by assigning them with (ii) it prevents unwanted information loss and incoherent outcomes.
the symbol ‘‘*". There are several existing studies those effectively dealt What is more, its validity is established through some axioms. Some
with 3WDM problems in IIS without compromising the originality of arithmetic operations and a possibility degree measure of IVF-CN are
data such as Liu et al. (2016), Xu et al. (2020) and Yang, Li et al. (2020). also developed.
However, their models only provided classification of objects but not Since some missing values are involved in an IIS. Therefore, we may
the ranking. Not but what, in practical problems, only envisaging the not aware of the exact illustration of these objects. In that case, the
classification of objects produces further difficulties for decision maker concept of equivalence relation cannot be applicable. Therefore, several
(DM) to make final decision. For example, if there are three sustainable existing studies have been proposed and explored by researchers on var-
community-based tourism (CBT) locations in the positive domain, then ious types of non-equivalence relations such as probability dominance
it is not possible for DM to select an optimal CBT. However, the studies relation (Wang et al., 2022c), partial-overall dominance relation (Yang,
those can provide the ranking solution alongside the classification in Deng et al., 2020), characteristic-based dominance relation (Du & Hu,
a 3WMADM model under IIS is rare. Pursuant to the discussions, a 2016), similarity relation (Zhan et al., 2021), etc. Consequently, in
3WMADM model under IIS is fabricated in this study to produce the this study, we propose a possibility dominance relation in an IVF-CN
solutions concerning the classification and ranking of the objects. incomplete information system (IVF-CNIIS) on the basis of possibility
Often we see that DM faces various inconveniences in order to degree of IVF-CN. The proposed possibility dominance relation is more
handle the ambiguities those are presented in complex decision making relaxed and effectively distinguish the objects based on the degree of
problems. Therefore, in the recent decades, a rapid improvement has superiorities. Subsequently, the possibility dominance class and set are
been noticed in handling uncertainty in decision making problems, also developed in virtue of the possibility dominance relation.
such as neutrosophic number (Giri & Roy, 2022), intuitionistic fuzzy The key point of 3WMADM is to construct loss functions for esti-
number (IFN) (Mondal et al., 2021), type-2 intuitionistic fuzzy (Das mating the risks of taking decisions. There are mainly two categories
et al., 2020), uncertain-random (Mondal & Roy, 2021), interval type- of framework, available in the previous studies. In the first category,
2 Pythagorean fuzzy (Mondal & Roy, 2022), fuzzy-rough (Roy et al., the loss functions are subjectively provided by the experts, such as Xin
2019), etc. In the matter of 3WMADM, although it is an efficient et al. (2021) and Zhang and Ma (2020), and in the other category, the
tool to reduce risks in decision making procedure, it still needs to loss functions are determined with the help of available information
improve by considering several ambiguities. Therefore, varieties of such as Gao et al. (2020) and Jia and Liu (2019). However, in the loss
uncertainties were implemented in existing 3WMADM models such as functions-based 3WDM models, alternatives with higher utility values
fuzzy set (Li, Xu et al., 2021; Zhan et al., 2021), IFN (Gao et al., 2020; could be missed. In regard of this fact, recently a new angle, namely
Jia & Liu, 2021), Pythagorean fuzzy number (PFN) (Liang et al., 2018), utility-based 3WDM models (Wang, Li, Qian et al., 2020; Zhan et al.,
interval type-2 fuzzy (Liang et al., 2019; Wang, Li, Qian et al., 2020), 2021) have been added to the research of 3WDM theory. Furthermore,
hesitant fuzzy (Lei et al., 2020; Wang et al., 2022a), etc. However, to encounter the behaviours of DM, several scientific behavioural the-
there are certain limitations of the previous-mentioned uncertainties ories such as prospect theory (Liang et al., 2019; Wang et al., 2005),
viz. fuzzy and interval type-2 fuzzy do not consider the membership regret theory (Wang, Li, Qian et al., 2020; Wang et al., 2022b; Zhu, Ma,
degrees, and IFN and PFN provide a limited scope for expression. For Zhan and Yao, 2022), etc. have been developed in 3WDM. Also, it is
instance, the belongingness and non-belongingness of an evaluation noticed in some existing works (Chorus, 2012) that the regret theory is
value are gathered from several experts as 0.6 to 0.8, and 0.5 to more advantageous than prospect or cumulative prospect theory. What
0.7, respectively. Therefore, the evaluation value cannot be illustrated is more, the subjective loss function and aggregation of loss functions
by existing uncertainties. In contrast, utilizing interval-valued Fer- may result in information distortion, and sometimes fail to provide
matean fuzzy number (IVFFN), the evaluation value can be depicted as satisfactory results. Thus, based on the above discussion and following
([0.6, 0.8], [0.5, 0.7]). Looking at these merits, IVFFN has gained notice- the work of Zhan et al. (2021), we develop the regret-based relative
able attention recently. For example, Jeevaraj (2021) exploited IVFFN utility function with the aid of regret theory. Therefore, the developed
in MADM problems. Rani and Mishra (2022) proposed IVFFN weighted relative utility function not only prevents subjective risks, but also
average operator (IVFFWAO) and IVFFN weighted geometric operator reflects DM’s regret and rejoice.
(IVFFWGO), and applied those in decision-making problems. Rani et al. Another crucial factor of 3WMADM is how to estimate the con-
(2022) developed Einstein operators on IVFFN and implemented those ditional probability. In a large number of studies, the conditional
in MADM problem. Accordingly, we express the evaluation information probability was calculated directly by adopting relative closeness ob-
via IVFFNs in this study. tained from distance-based MADM methods such as technique for order
However, modern decision making problems become more incom- of preference by similarity to ideal solution (Gao et al., 2020; Liang
modious when various parameters are shapely and it hinges on the et al., 2018), grey relational analysis (Liu, Wang et al., 2020), etc.
behaviours of the evaluators (Liu, Malik et al., 2020). Based on this Nevertheless, in their models, the attributes must be of benefit and
fact, Zhao (1989) introduced set pair analysis (SPA) theory by fusing cost types. But, in some real-life decision making problems, the above
dialectical reasoning and mathematical tools, which describes certainty assumption may not hold for some attributes which are neither benefit
and uncertainty in a single structure. The primary element of SPA type nor cost type. Therefore, for this type of problems, it is not
theory is the connection number (CN) which is a form of three angles, logical to directly fuse the evaluation values of all attributes and to
namely, 𝑎 ∶ identity, 𝑏 ∶ discrepancy, and 𝑐 ∶ contrary, with the adopt the relative closeness as the conditional probability. On the other
condition 𝑎 + 𝑏 + 𝑐 = 1. Recently, SPA theory has been utilized by hand, some previous 3WMADM models utilized subjective conditional
several researchers such as Garg and Kumar (2020), Kumar and Chen probability (Jia & Liu, 2019), which created further vagueness and risks
(2021) and Liu, Malik et al. (2020), etc. The existing studies restricted in decision framework. In order to calculate the conditional probabil-
the scope for CN within the condition 𝑎 + 𝑏 + 𝑐 = 1. However, in ity, one of the essentials, which needs to be derived is the decision
many real-life problems, the condition does not hold and may be attribute. Whereas an IMADM may not possess a decision attribute.
𝑎 + 𝑏 + 𝑐 > 1 (Garg, 2021). Also, in MADM problems, aggregation of However, most of the previous studies either ignored it (Huang &
2
A. Mondal et al. Expert Systems With Applications 211 (2023) 118688
Zhan, 2021; Wang et al., 2022a; Zhu, Ma, and Zhan, 2022) or derived (i) The vagueness in evaluation information is limned by IVFFN,
subjectively (Yang, Deng et al., 2020; Zhan et al., 2021). In contrast we which provides enough scope for experts to express their opin-
propose a new method to derive objective pre-decision that has two ions. Furthermore, to prevent incoherent results, we introduce
purposes: (i) to provide a basic notion about the upcoming decision a novel uncertainty, namely IVF-CN by combining IVFFN and
procedure, and (ii) to deduce a fuzzy decision attribute (or fuzzy state SPA theory. Since IVF-CN provides three regions to describe the
set). Hence, we calculate the conditional probability by making full use uncertainty, it is very correlated to 3WDM theory which also
of the evaluation information in light of fuzzy decision attribute and delivers three domains.
possibility dominance set. (ii) Most of the relations in existing 3WMADM models are defined
under crisp environment. On the other hand, we define a more
relaxed possibility dominance relation under uncertain environ-
1.1. Motivations
ment in IVF-CNIIS, which can effectively differentiate the objects
based on their superiorities.
Based on the above discussions, we summarize the main motivations (iii) In order to reduce the decision-based risk, occurred due to
to develop this study as follows: considering subjective factors, we offer a new method based on
the original evaluation information to derive the objective fuzzy
(i) Most of the previous studies on 3WMADM only encountered
state set.
complete MADM problems (Liang et al., 2018; Wang, Li, Qian
(iv) In contrast, we compute different relative utility functions for
et al., 2020; Wang, Li, Zhang et al., 2020; Wang et al., 2022a).
each alternative by utilizing the fuzzy state set and regret the-
However, in practical situations, it may not be always possible
ory. Therefore, the proposed model not only cut down the risk
to gather all information in a decision making problem. Fur-
of subjective loss functions, but also takes into account the
thermore, the majority of 3WMADM models in IISs considered
psychological and risk attitude of DM.
the vagueness or ambiguity in evaluation information as fuzzy
(v) We propose a new method to estimate the objective conditional
set (Liu et al., 2016; Yang, Li et al., 2020; Zhan et al., 2021), or
probability with the assistance of fuzzy state set and possibility
as IFN (Xin et al., 2021), and some of them ignored it (Ye et al.,
dominance set.
2021). These models can deal with the practical problems up to (vi) The proposed 3WMADM model does not tamper with the orig-
a certain limit. inal information as it neither fills nor ignores the missing data.
(ii) In previous studies on 3WMADM, the fuzzy state set was either Also, it effectively provides the classification as well as ordering
ignored (Huang & Zhan, 2021; Wang et al., 2022a; Xin et al., of all alternatives.
2021; Zhu, Ma, and Zhan, 2022), or subjectively given (Ye et al.,
2021; Zhan et al., 2021). However, it is one of the indispensable The paper is farther laid out as follows: Section 2 introduces some
factors in a 3WDM model. Thus, inappropriate results or severe prerequisites concepts. Section 3 introduces the IVF-CN and the related
risk can be occurred by ignoring or subjectively providing it. concepts. Section 4 develops the possibility dominance relation and
(iii) A good number of previous studies on 3WMADM has been the regarding notions. The proposed 3WMADM model is illustrated
constructed only information table and overlooked the loss func- in Section 5. Section 6 demonstrates the application of the proposed
tions (Yang, Li et al., 2020) or used the same loss functions 3WMADM model in two real-life case studies. Section 7 presents the
for all alternatives (Zhang & Ma, 2020). However, in real-world comparative analysis with several existing studies to elucidate the va-
actual MADM problems, each alternative should have different lidity and superiority of the proposed model. The experimental analysis
loss functions. Again, some of the researchers adopted subjective is conducted in Section 8 to investigate the effect of various parameters
loss/utility functions provided by the experts (Li, Wang et al., to the model. Finally, Section 9 exhibits the conclusion and point out
2021; Wang, Li, Qian et al., 2020) or fused the loss functions some future scopes.
using aggregation operators (Li, Xu et al., 2021). The former case
is too subjective and the later one may lead to information loss. 2. Prerequisites
(iv) A lot of previous studies utilized distance-based methods to
determine the conditional probability (Gao et al., 2020; Liang In this section, we provide the fundamentals regarding the proposed
et al., 2018) or deemed subjective conditional probability (Jia & study.
Liu, 2019, 2021) or ignored it (Yang, Li et al., 2020). However,
the distance-based methods cannot be applied in every realistic Definition 2.1 (Jeevaraj, 2021). An interval-valued
{( [ Fermaten] [fuzzy set
situations and ignoring it or adopting subjective conditional on a universal set 𝑋 is defined as = 𝑥, 𝜇𝐿 (𝑥), 𝜇𝑈 (𝑥) , 𝜈𝐿 (𝑥),
]) }
probability may not produce satisfactory outcomes because of 𝜈𝑈 (𝑥) |𝑥 ∈ 𝑋 in which [𝜇𝐿 (𝑥), 𝜇𝑈 (𝑥)]
subjective risks. [ ]
⊆ [0, 1] and 𝜈𝐿 (𝑥), 𝜈𝑈 (𝑥) ⊆ [0, 1] are the interval-valued membership
(v) Some of the existing 3WMADM models in IMADM environment
filled the unknown data and operated as complete MADM prob- and non-membership degrees of 𝑥 satisfying 0 ≤ 𝜇𝐿 (𝑥) ≤ 𝜇𝑈 (𝑥) ≤
( )3 ( )3
lems in decision making process (Xin et al., 2021; Ye et al., 1, 0 ≤ 𝜈𝐿 (𝑥) ≤ 𝜈𝑈 (𝑥) ≤ 1 and 𝜇𝑈 (𝑥) + 𝜈𝑈 (𝑥) ≤ 1.
2021). This can lead to variations in original information, the
Correspondingly,
[√ the interval-valued hesitancy degree of 𝑥 is defined
]
result of which may not truly reflect the practical situation. ( )3 ( )3 √ ( )3 ( )3
3 3
Moreover, some of the previous 3WMADM models in an IIS as 1 − 𝜇𝑈 (𝑥) − 𝜈𝑈 (𝑥) , 1 − 𝜇𝐿 (𝑥) − 𝜈𝐿 (𝑥) . For
only deduced the classification and passed by the ranking of ([ ] [ ])
alternatives (Liang et al., 2018; Liu et al., 2016). This cannot convenience, an IVFFN is expressed as 𝜇𝐿 , 𝜇𝑈 , 𝜈𝐿 , 𝜈𝑈 .
be very fruitful for DM in practical.
Definition 2.2 (Wang et al., 2005). Let 𝐼𝑘 = [𝑖− , 𝑖+ ] (𝑘 = 1, 2) be two
𝑘 𝑘
intervals, then the preference degree of 𝐼1 ≻ 𝐼2 is given as
1.2. Contributions
( ) ( )
max 0, 𝑖+
1
− 𝑖−
2
− max 0, 𝑖−1
− 𝑖+
2
𝑃 (𝐼1 ≻ 𝐼2 ) = (+ ) ( ) .
In response to the motivations illustrated in the previous subsection, 𝑖1 − 𝑖−
1
+ 𝑖+2
− 𝑖−
2
we deliver the following contributions through this study.
3
A. Mondal et al. Expert Systems With Applications 211 (2023) 118688
Definition 2.3 (Zhao, 1989). Let (, ) be a set pair between two Table 1
The utility functions for different actions.
correlated sets and for a problem with features, in which
features are the same between and , features are conflicting ¬
4
A. Mondal et al. Expert Systems With Applications 211 (2023) 118688
Fig. 1. The utility function (a) and regret–rejoice function (b) for different values of 𝜃 and 𝛿, respectively.
Definition
([ 3.2.] [ The ]) IVF-CN corresponding to an IVFFN
= 𝜇𝐿 , 𝜇𝑈 , 𝜈𝐿 , 𝜈𝑈 is defined as: 𝜓 = 𝑎 + 𝑏 𝑖 + 𝑐 𝑗, in
Again, according to Zhan et al. (2021), the relative utility functions are
presented in Table 2. In Table 2, 𝜔 ̂𝑃 𝑃 = 𝜔𝑃 𝑃 −𝜔𝑁𝑃 , 𝜔
̂𝐵𝑃 = 𝜎 𝜔 ̂𝐵𝑁 =
̂𝑃 𝑃 , 𝜔 which the identity (𝑎 ), discrepancy (𝑏 ) and contrary (𝑐 ) degrees are
̂𝑁𝑁 and 𝜔
𝜎𝜔 ̂𝑁𝑁 = 𝜔𝑁𝑁 −𝜔𝑃 𝑁 , where 𝜎 is the utility pursuit coefficient. described as follows:
√( )( )
Therefore, the new decision rules are illustrated as follows: 3
𝜇3 + 𝜇3 4 − 2𝜈3 − 2𝜈3
(P1′ ) If Pr( ∣ [𝑜]𝑅 ) ≥ 𝛼
̂, then decide 𝑜 ∈ 𝑃 𝑂𝑆(), 𝐿 𝑈 𝐿 𝑈
𝑎 = ,
(B1′ ) If 𝛽̂ ≤Pr( ∣ [𝑜]𝑅 ) ≤ 𝛼
̂, then decide 𝑜 ∈ 𝐵𝑁𝐷(), 2
̂ then decide 𝑜 ∈ 𝑁𝐸𝐺(). √
(N1′ ) If Pr( ∣ [𝑜]𝑅 ) ≤ 𝛽, 3 3 3 3
3 1 + (1 − 𝜇 − 𝜇 )(1 − 𝜈 − 𝜈 )
Here, 𝑏 = 𝐿 𝑈 𝐿 𝑈
,
2
𝜔
̂𝐵𝑁 𝜎𝜔̂𝑁𝑁 √( )( )
̂=
𝛼 = . (2.7)
̂𝐵𝑁 + (𝜔
𝜔 ̂𝑃 𝑃 − 𝜔̂𝐵𝑃 ) ̂𝑁𝑁 + (1 − 𝜎)𝜔
𝜎𝜔 ̂𝑃 𝑃
3
𝜈3 + 𝜈3 4 − 2𝜇3 − 2𝜇3
𝐿 𝑈 𝐿 𝑈
̂𝑁𝑁 − 𝜔
𝜔 ̂𝐵𝑁 (1 − 𝜎)𝜔
̂𝑁𝑁 𝑐 = .
𝛽̂ = = . (2.8) 2
(𝜔
̂𝑁𝑁 − 𝜔̂𝐵𝑁 ) + 𝜔
̂𝐵𝑃 (1 − 𝜎)𝜔
̂𝑁𝑁 + 𝜎 𝜔
̂𝑃 𝑃
Also, the expected utility values for taking different actions are illus-
trated as follows: Remark 3.1. Since 𝑎 is a function of membership degrees and
complement of non-membership degrees, therefore, it expresses the
̂𝑃 𝑃 Pr( ∣ [𝑜]𝑅 ),
𝛤 (𝜏𝑃 ∣ [𝑜]𝑅 ) = 𝜔
( ) identity degree. Similarly, 𝑐 is a function of non-membership de-
𝛤 (𝜏𝐵 ∣ [𝑜]𝑅 ) = 𝜎 𝜔 ̂𝑃 𝑃 Pr( ∣ [𝑜]𝑅 ) + 𝜔
̂𝑁𝑁 𝑃 𝑟(¬ ∣ [𝑜]𝑅 ) , grees and complement of membership degrees, therefore, it expresses
̂𝑁𝑁 Pr(¬ ∣ [𝑜]𝑅 ).
𝛤 (𝜏𝑁 ∣ [𝑜]𝑅 ) = 𝜔 the contrary degree. Also, 𝑏 is a function of complements of both
membership and non-membership degrees, therefore, it expresses the
3. The IVF-CN discrepancy degree. Hence, 𝑎 , 𝑏 and 𝑐 are logical.
Definition 3.1. The CN for a set pair (, ) is defined as 0≤ 2(𝜈3 + 𝜈3 ) ≤ 4
𝐿 𝑈
5
A. Mondal et al. Expert Systems With Applications 211 (2023) 118688
Hence, the IVF-CN defined in Definition 3.2 is valid. 𝐼𝑉 𝐹 -𝐶𝑁𝑊 𝐺𝑤 (𝜓1 , 𝜓2 , … , 𝜓𝑛 )
( )1
Definition 3.3. Let 𝜓𝑘 = 𝑎𝑘 + 𝑏𝑘 𝑖 + 𝑐𝑘 𝑗 (𝑘 = 1, 2) be two IVF-CNs, ∏
𝑛 ∏
𝑛
3
= (𝑎𝑝 )𝑤𝑝 + 1− (1 − 𝑏3 )𝑤𝑝 𝑖
then 𝑝
𝑝=1 𝑝=1
(i) 𝜓1 ≈ 𝜓2 if and only if 𝑎1 = 𝑎2 and 𝑐1 = 𝑐2 . ( 𝑛 (
)1
∏ )𝑤𝑝 𝑛 (
∏ )𝑤𝑝 3
(ii) 𝜓1 ⪯ 𝜓2 if and only if 𝑎1 ≤ 𝑎2 and 𝑐1 ≥ 𝑐2 . + 𝑎3 + 𝑐3 − 𝑎3 𝑗.
𝑝 𝑝 𝑝
𝑝=1 𝑝=1
Definition 3.4. Let 𝜓𝑘 = 𝑎𝑘 + 𝑏𝑘 𝑖 + 𝑐𝑘 𝑗, 𝑘 = 1, 2 be two IVF-CNs,
then the multiplication and exponent operations
(√ are defined as: ) Proof. Based on (ii) of Definition 3.4, we have
(i) 𝜓1 ⊗ 𝜓2 = 𝑎1 𝑎2 + 3 1 − (1 − 𝑏3 )(1 − 𝑏3 ) 𝑖 +
(√ ) 1 2
𝑤
( ( )𝑤𝑝 ) 1 (( )𝑤 𝑝
3 (𝑎3 + 𝑐 3 )(𝑎3 + 𝑐 3 ) − 𝑎3 𝑎3 𝑗. 𝜓 𝑝 = (𝑎𝑝 )𝑤𝑝 + 1 − 1 − 𝑏3 3 𝑖 + 𝑎3 + 𝑐 3
1 1 2 2 1 2
(√ ) (√ ) 𝑝 𝑝 𝑝 𝑝
𝜓1 ⊗ 𝜓2 = ( 𝑛 (
)1
(√ ) ∏ )𝑤 𝑝 ∏𝑛 ( )𝑤 𝑝 3
𝑎1 𝑎2 + 3 𝑎3 𝑏3 + 𝑏3 𝑎3 + 𝑏3 𝑏3 + 𝑏3 𝑐3 + 𝑐3 𝑏3 𝑖 + 𝑎3 + 𝑐3 − 𝑎3 𝑗.
1 2 1 2 1 2 1 2 1 2 𝑝 𝑝 𝑝
(√ ) 𝑝=1 𝑝=1
+ 3 𝑎3 𝑐3 + 𝑐3 𝑎3 + 𝑐3 𝑐3 𝑗 Hence, Theorem 3.2 is proved.
1 2 1 2 1 2
(√ )
= 𝑎1 𝑎2 + 3
𝑏 (𝑎 + 𝑐 ) + 𝑏3 (𝑎3 + 𝑐3 ) + 𝑏3 𝑏3 𝑖+
3 3 3
1 2 2 2 1 1 1 2 Definition 3.6. Let 𝜓𝑘 = 𝑎𝑘 + 𝑏𝑘 𝑖 + 𝑐𝑘 𝑗 (𝑘 = 1, 2) be two IVF-CNs,
(√ ) ( )
3
𝑎3 𝑐3 + 𝑐3 𝑎3 + 𝑐3 𝑐3 + 𝑎3 𝑎3 − 𝑎3 𝑎3 𝑗. then the possibility degree 𝑝𝑜𝑠 𝜓1 ⪰ 𝜓2 of (𝜓1 ⪰ 𝜓2 ) is defined
1 2 1 2 1 2 1 2 1 2
6
A. Mondal et al. Expert Systems With Applications 211 (2023) 118688
Proof. Since, in the IVF-CN 𝜓 = 𝑎 + 𝑏 𝑖 + 𝑐 𝑗, the larger value of de- [𝑜𝑝 ]≥ ∣ 𝑜𝑝 ∈ 𝑂 constitutes a covering of 𝑂.
𝑎𝑗
gree of discrepancy 𝑏 means the bigger possible change scope for the
degrees of identity and contrary as 𝑏3 = 1−𝑎3 −𝑐3 . Thus,
[ the√degrees of
] Definition 4.3. Let ∗ = (𝑂, 𝐴, 𝑉 ∗ , 𝑓 ) be an IVF-CNIIS and 𝐵 ⊆ 𝐴, then
identity and contrary can be expressed as the intervals 𝑎 , 𝑎3 + 𝑏3
3
the possibility dominance set of 𝑜𝑝 ∈ 𝑂 is defined as:
[ √ ] ⋂
and 𝑐 , 3 𝑏3 + 𝑐3 , respectively. Therefore, the comparison of two ≥ (𝑜 ) =
𝐵 𝑝
[𝑜𝑝 ]≥ .
𝑎𝑗
𝑎𝑗 ∈𝐵
IVF-CNs 𝜓1 and 𝜓2 can be made by using the preference degree for-
[ √ ] [ √ ]
mula for the intervals 𝑎1 , 3 𝑎3 + 𝑏3 and 𝑎2 , 3 𝑎3 + 𝑏3 . Then, Definition 4.4. Let ∗ = (𝑂, 𝐴, 𝑉 ∗ , 𝑓 ) be an IVF-CNIIS and 𝐵 ⊆ 𝐴,
1 1 2 2
then the binary judgement value of 𝑜𝑝 and 𝑜𝑞 under the attribute 𝑎𝑗 is
on the basis of Definition 2.2, the possibility degree of 𝜓1 ⪰ 𝜓2 is
defined as:
defined as:
( ) ⎧
𝑝𝑜𝑠 𝜓1 ⪰ 𝜓2 ⎪1, 𝑜𝑝 ∈ [𝑜𝑞 ]≥ ,
𝑎𝑗
𝑙𝑎𝑗 (𝑜𝑝 , 𝑜𝑞 ) = ⎨ (4.12)
( √ ) ( √ ) ⎪0, otherwise,
max 0, 3 𝑎3 + 𝑏3 − 𝑎2 − max 0, 𝑎1 − 3 𝑎3 + 𝑏3 ⎩
1 1 2 2
= √ √ . where [𝑜𝑞 ]≥ is the possibility dominance class of 𝑜𝑞 relative to the
3 𝑎3 + 𝑏 3 − 𝑎 3 𝑎3 + 𝑏 3 − 𝑎
1 +
𝑎𝑗
2
1 1 2 2 attribute 𝑎𝑗 .
Also, since 𝑎3 + 𝑏3 + 𝑐3 = 1, then we derive
1 1 1
( √ ) ( √ ) Definition 4.5. Let ∗ = (𝑂, 𝐴, 𝑉 ∗ , 𝑓 ) be an IVF-CNIIS and 𝐵 ⊆ 𝐴, then
3 1 − 𝑐3 − 𝑎 3 1 − 𝑐3
( ) max 0, 1 2 − max 0, 𝑎1 − 2
the overall judgement value between two objects 𝑜𝑝 , 𝑜𝑞 ∈ 𝑂 is defined
𝑝𝑜𝑠 𝜓1 ⪰ 𝜓2 = √ √ . as:
3 𝑎3 + 𝑏 3 − 𝑎
1 +
3 𝑎3 + 𝑏 3 − 𝑎
2 ∑
1 1 2 2 𝐵 (𝑜𝑝 , 𝑜𝑞 ) = 𝑤𝑗 𝑙𝑎𝑗 (𝑜𝑝 , 𝑜𝑞 ), (4.13)
Thus, the result in Eq. (3.10) is proved. Again, when 𝑏1 = 0 and 𝑎𝑗 ∈𝐵
𝑏2 = 0, then the intervals reduce to real numbers. Hence, the results ∑
where 𝑤𝑗 is the weight of 𝑗th attribute such that 𝑗 𝑤𝑗 = 1. Based on
in Eq. (3.11) are evident. the overall judgement values among objects, a judgement matrix can
be constructed with respect to 𝐵, from which the preference degrees of
Theorem 3.3. The possibility degree of IVF-CNs defined in Definition 3.6 the objects are calculated as follows:
obeys the following properties: ∑
( ) 1
𝐷𝐵 (𝑜𝑝 ) = (𝑜 , 𝑜 ), ∀ 𝑜𝑝 ∈ 𝑂. (4.14)
(i) 0 ≤ 𝑝𝑜𝑠 𝜓1 ⪰ 𝜓2 ≤ 1. ∣ 𝑂 ∣ −1 𝑝≠𝑞 𝐵 𝑝 𝑞
( )
(ii) 𝑝𝑜𝑠 𝜓1 ⪰ 𝜓2 = 0.5 if 𝜓1 ≈ 𝜓2 . It is evident that 0 ≤ 𝐷𝐵 (𝑜𝑝 ) ≤ 1.
( ) ( )
(iii) 𝑝𝑜𝑠 𝜓1 ⪰ 𝜓2 + 𝑝𝑜𝑠 𝜓2 ⪰ 𝜓1 = 1.
( ) 5. A novel 3WMADM model under IIS
(iv) 𝑝𝑜𝑠 𝜓1 ⪰ 𝜓2 = 1 ⇔ 𝑐13 + 𝑎32 ≤ 1 and 𝑎31 + 𝑐23 ≥ 1.
( )
(v) 𝑝𝑜𝑠 𝜓1 ⪰ 𝜓2 = 0 ⇔ 𝑐13 + 𝑎32 ≥ 1 and 𝑎31 + 𝑐23 ≤ 1. In this section, we propose an MADM model which turns the tra-
ditional two-way MADM problem into a 3WMADM problem with the
aid of information and utility function tables under IVF-CNIIS. The
Proof. All cases can be proven by using the properties of IVF-CN. mathematical description of the MADM is as follows:
{ }
Let 𝑂 = 𝑜1 , 𝑜2 , … , 𝑜𝑚 be the set of 𝑚 alternatives (or objects)
4. Possibility dominance relation in IVF-CNIIS which have to be evaluated with respect to the set of 𝑛 attributes,
{ }
𝐴 = 𝑎1 , 𝑎2 , … , 𝑎𝑛 in an MADM problem. Also, let 𝑊 = (𝑤1 , 𝑤2 , … , 𝑤𝑛 )
In this section, the dominance relation, dominance class and their be the weight vector of the set of attributes and 𝑜𝑝𝑗 represents the
corresponding concepts are introduced based on the possibility degree assessment value of the 𝑝th alternative with respect to the 𝑗th attribute.
of IVF-CN to deal with unknown information involved in an IVF-CNIIS. If there exist some 𝑝 ∈ {1, 2, … , 𝑚} and 𝑗 ∈ {1, 2, … , 𝑛} such that
7
A. Mondal et al. Expert Systems With Applications 211 (2023) 118688
𝑜𝑝𝑗 =∗, then the MADM problem is considered as an IMADM problem. Proof. Since (𝑜𝑡 ) + ¬(𝑜𝑡 ) = 1, then form Eqs. (5.15)–(5.16) we have
Furthermore, when the evaluation values are represented by IVFFN and ∑ ∑
(𝑜𝑡 ) + ¬(𝑜𝑡 )
IVF-CN, then the MADM and IMADM can be regarded as the IVFIIS and
( ) ( ) 𝑜𝑡 ∈≥ (𝑜 )
𝐴 𝑝
𝑜𝑡 ∈≥ (𝑜 )
𝐴 𝑝
IVF-CNIIS, respectively. Pr ∣ 𝑜𝑝 + Pr ¬ ∣ 𝑜𝑝 =
| ≥ |
|𝐴 (𝑜𝑝 )|
∑ ( | |
5.1. Pre-decision for IMADM )
(𝑜𝑡 ) + ¬(𝑜𝑡 )
𝑜𝑡 ∈≥ (𝑜 )
𝐴 𝑝
As described in introduction, one of the necessary factors for esti- =
| ≥ |
mating the conditional probability is decision attribute. Also, it delivers |𝐴 (𝑜𝑝 )|
| |
a preliminary idea to DM about the classification of the objects, which | ≥ |
|𝐴 (𝑜𝑝 )|
can be regarded as pre-classification. According to Zhan et al. (2021), = | | = 1.
| ≥ |
the decision obtained in pre-decision has a definite influence in final |𝐴 (𝑜𝑝 )|
| |
decision. Thus, the pre-decision has mainly two objectives, namely, it
provides a pre-classification, based on which, we can argue about the 5.3. Regret-based utility function
rationality of final decision, and it generates fuzzy decision attribute
which is one of the prerequisites to calculate the conditional prob- Utility function is the another key point of the 3WMADM which
ability. In what follows, we propose the pre-decision for an IMADM is necessary for calculating the values of decision thresholds. Utility
problem. functions can directly be obtained from DM, which may provide infor-
mation loss and produce undesired outcomes. Consequently, motivated
Definition 5.1. Suppose in an IMADM problem, the preference degree by the idea of Zhan et al. (2021), we propose a new procedure to
of each alternative is defined by Eq. (4.14), then the state of a fuzzy estimate the relative utility functions with the assistance of regret
𝜆 𝜆 𝜆 theory. Since 𝜔 ̂𝑃 𝑃 represents the utility for choosing an alternative,
concept ‘‘good alternative’’ is defined as = 1 + 2 + ⋯ + 𝑚 , where
𝑜1 𝑜2 𝑜𝑚 then it should involve DM’s behaviours towards regret and risk to make
𝜆𝑝 is the corresponding membership degree of 𝑝th alternative and more realistic decisions. Sometimes DM may feel regret for choosing a
𝜆𝑝 = 𝐷𝐴 (𝑜𝑝 ), ∀ 𝑝. Therefore, the alternatives can be ranked according certain alternative instead of the optimal alternative depending upon
to the membership degree of each alternative, which is said to be a the decision making situation. Utilizing 𝜆𝑝 in 𝜔
̂𝑃 𝑃 , we cannot reflect the
pre-decision for an IMADM. riskiness of choosing the alternative 𝑜𝑝 . Hence, in light of 𝜆𝑝 and regret
theory, we give a new calculation procedure of 𝜔 ̂𝑃 𝑃 in the following
It is conspicuous that, only ranking of the alternatives can be
definition.
obtained from the pre-decision, but not the classification of the alterna-
tives. Therefore, in the following, we develop a 3WMADM model under
Definition 5.2. Suppose in an IVF-CNIIS, the state is given as a fuzzy
IVF-CNIIS to categorize the alternatives. 𝜆 𝜆 𝜆
set as = 1 + 2 +⋯+ 𝑚 , where 𝜆𝑝 = 𝐷𝐴 (𝑜𝑝 ) is the preference degree
𝑜1 𝑜2 𝑜𝑚
5.2. Estimating conditional probability of 𝑝th alternative, then the relative utility function of 𝑝th alternative
incurred for adopting the action 𝜏𝑃 under the state is defined as:
As mentioned in introduction, in an IIS, the notion of equivalence {
𝑝 𝑢(𝜆𝑝 ) + 𝑟(𝑢(𝜆𝑝 ) − 𝑢(𝜆∗ )), 𝜆𝑝 ≠ 0,
relation is no longer applicable. Therefore, in light of the possibility ̂𝑃 𝑃 =
𝜔 (5.17)
0, 𝜆𝑝 = 0,
dominance relation and fuzzy state set, the conditional probabilities of
the alternatives are estimated by fully utilizing the information related 1 − 𝑒−𝜃𝜆𝑝 ∗
to each attribute in the frame of IVF-CNIIS. For this purpose, we define where 𝑢(𝜆) = and 𝑟(𝑢(𝜆𝑝 ) − 𝑢(𝜆∗ )) = 1 − 𝑒−𝛿(𝑢(𝜆𝑝 )−𝑢(𝜆 )) in which
∗ 𝜃
the set of states as: 𝛺 = {, ¬} such that the fuzzy set connotes a 𝜆 = max𝑝 𝜆𝑝 , and, 𝜃 and 𝛿 are the risk and regret aversion coefficients
fuzzy concept of ‘‘good alternatives’’ and the complement ¬ connotes of DM, respectively.
a fuzzy concept of ‘‘bad alternatives’’. Furthermore, the fuzzy set Again, DM may also feel regret for rejecting a certain alternative
𝜆 𝜆 𝜆
and its complement ¬ are given as = 1 + 2 + ⋯ + 𝑚 and instead of the worst alternative. Since 𝜔̂𝑁𝑁 denotes the utility for
𝑜1 𝑜2 𝑜𝑚
1 − 𝜆1 1 − 𝜆2 1 − 𝜆𝑚 rejecting an alternative. Thus, we provide the following definition for
¬ = + +⋯+ , respectively. Then, the conditional computing 𝜔 ̂𝑁𝑁 .
𝑜1 𝑜2 𝑜𝑚
probabilities of each alternative are defined as:
∑ Definition 5.3. Suppose in an IVF-CNIIS, the state ¬ is given as a
(𝑜𝑡 )
𝜆 𝜆 𝜆
( ) 𝑜𝑡 ∈≥ (𝑜 )
𝐴 𝑝
fuzzy set as = 1 + 2 + ⋯ + 𝑚 , where 𝜆𝑝 = 1 − 𝜆𝑝 , then the relative
Pr ∣ 𝑜𝑝 = , ∀ 𝑝 ∈ {1, 2, … , 𝑚}, (5.15) 𝑜1 𝑜2 𝑜𝑚
| ≥ |
|𝐴 (𝑜𝑝 )| utility function of 𝑝th alternative caused for taking the action 𝜏𝑁 under
|∑ |
the state ¬ is defined as:
¬(𝑜𝑡 ) { ∗
( ) 𝑜𝑡 ∈≥ (𝑜
𝐴 𝑝
) 𝑝 𝑢(𝜆𝑝 ) + 𝑟(𝑢(𝜆𝑝 ) − 𝑢(𝜆 )), 𝜆𝑝 ≠ 0,
Pr ¬ ∣ 𝑜𝑝 = , ∀ 𝑝 ∈ {1, 2, … , 𝑚}, (5.16) ̂𝑁𝑁 =
𝜔 (5.18)
| ≥ | 0, 𝜆𝑝 = 0,
|𝐴 (𝑜𝑝 )|
| |
where (𝑜𝑡 ) = 𝜆𝑡 = 𝐷𝐴 (𝑜𝑡 ) and ¬(𝑜𝑡 ) = 1 − 𝜆𝑡 = 1 − 𝐷𝐴 (𝑜𝑡 ) and ≥ (𝑜 ) 1 − 𝑒−𝜃𝜆𝑝 ∗ ∗
𝐴 𝑝 where 𝑢(𝜆) = and 𝑟(𝑢(𝜆𝑝 ) − 𝑢(𝜆 )) = 1 − 𝑒−𝛿(𝑢(𝜆𝑝 )−𝑢(𝜆 )) in which
is the possibility dominance set of the alternative 𝑜𝑝 . In Eq. (5.15), ∗ 𝜃
( ) 𝜆 = max𝑝 𝜆𝑝 , and, 𝜃 and 𝛿 are the risk and regret aversion coefficients
Pr ∣ 𝑜𝑝 expresses the probability that the alternative 𝑜𝑝 belongs to
( ) of DM, respectively.
the state and in Eq. (5.16), Pr ¬ ∣ 𝑜𝑝 elicits the probability that
the alternative 𝑜𝑝 belongs to the state ¬.
̂𝑝𝑃 𝑃 ≤ 𝜔
Proposition 5.2. If 𝜆𝑝 ≤ 𝜆𝑞 , then 𝜔 ̂𝑞𝑃 𝑃 and 𝜔
̂𝑝𝑁𝑁 ≥ 𝜔
̂𝑞𝑁𝑁 .
8
A. Mondal et al. Expert Systems With Applications 211 (2023) 118688
Based on Proposition 5.2, it is clear that the proposed utility func- Table 3
tions will improve the capability of DM to distinguish alternatives. The relative utility functions for 𝑜𝑝 .
Hence, with the aid of regret-based utility functions and Table 2, the ¬
relative utility functions of the alternative 𝑎𝑝 are depicted in Table 3, 𝜏𝑃 ̂𝑝𝑃 𝑃
𝜔 0
in which 𝜔 ̂𝑝𝐵𝑃 = 𝜎 𝜔
̂𝑝𝑃 𝑃 and 𝜔
̂𝑝𝐵𝑁 = 𝜎 𝜔
̂𝑝𝑁𝑁 . It is evident that the larger 𝜏𝐵 ̂𝑝𝐵𝑃
𝜔 ̂𝑝𝐵𝑁
𝜔
the value of 𝜎, the greater the values of 𝜔 ̂𝑝𝐵𝑃 and 𝜔 ̂𝑝𝐵𝑁 are. Therefore, 𝜏𝑁 0 ̂𝑝𝑁𝑁
𝜔
Remark 5.1. The main differences of the proposed utility function and (i) If 𝜆𝑝 = 0, then 𝜔̂𝑝𝑃 𝑃 = 0. Therefore, from Eqs. (5.22)–(5.23) we
existing ones are as follows:
̂𝑝 = 𝛽̂𝑝 = 1, which implies that only the action of rejection
get 𝛼
• Zhan et al. (2021) proposed utility function by considering devia- can be adopted for the alternative 𝑎𝑝 .
tion of utility value of an alternative from the mean utility value. (ii) If 𝜆𝑝 = 1, then 𝜔̂𝑝𝑁𝑁 = 0. Therefore, from Eqs. (5.22)–(5.23) we
On the other hand, the proposed utility function is calculated by ̂𝑝 = 𝛽̂𝑝 = 0, which implies that only the action of acceptance
get 𝛼
considering deviation from the optimal utility value. can be adopted for the alternative 𝑎𝑝 .
• Wang, Li, Qian et al. (2020) computed utility function by using
subjective outcome matrix. In contrast, we utilize objective utility
value that obtained from fuzzy state set to derive utility function. Proposition 5.3. Assume 𝜆𝑝 ∈ (0, 1) and 𝜎 > 0, then
• Zhang et al. (2018) derived utility function based on traditional
̂𝑝 > 𝛽̂𝑝 if and only if 𝜎 > 0.5.
(i) 𝛼
utility function that cannot reflect DM’s behaviours regarding
risk and regret. Oppositely, the proposed utility function is de- (ii) When 𝜎 > 0.5 and the value of 𝜎 increases, the value of 𝛼
̂𝑝 increases
rived based on regret theory that involves DM’s risk and regret and the value of 𝛽̂𝑝 decreases.
attitudes.
Proof. Proof of this proposition can be viewed in Zhan et al. (2021).
5.4. The novel 3WMADM model
What is more, the associated utility of the alternative 𝑎𝑝 is calculated
Based on the discussions in Sections 5.2 and 5.3, this subsection as follows:
presents a novel 3WMADM model under IVFIIS. At first, the IVFFN ⎧𝛤 (𝜏𝑃 ∣ 𝑜𝑝 ), if 𝑜𝑝 ∈ 𝑃 𝑂𝑆(),
evaluation values are converted into IVF-CN evaluation values to trans- ⎪
𝛬(𝑎𝑝 ) = ⎨𝛤 (𝜏𝐵 ∣ 𝑜𝑝 ), if 𝑜𝑝 ∈ 𝐵𝑁𝐷(), (5.24)
form the IVFIIS into an IVF-CNIIS. Therefore, in light of the conditional ⎪
probability and the relative utility functions perused in Sections 5.2 and ⎩𝛤 (𝜏𝑁 ∣ 𝑜𝑝 ), if 𝑜𝑝 ∈ 𝑁𝐸𝐺(),
5.3, the expected utility values of the alternative 𝑎𝑝 with reference to where 𝛤 (𝜏𝑃 ∣ 𝑜𝑝 ), 𝛤 (𝜏𝐵 ∣ 𝑜𝑝 ) and 𝛤 (𝜏𝑁 ∣ 𝑜𝑝 ) are given in Eqs. (5.19),
three actions 𝜏∙ (∙ = 𝑃 , 𝐵, 𝑁) are elucidated as follows: (5.20) and (5.21), respectively. Since the priority of three regions is as:
( ) 𝑃 𝑂𝑆() ≻ 𝐵𝑁𝐷() ≻ 𝑁𝐸𝐺(). Therefore, the alternatives are ranked
̂𝑝𝑃 𝑃 Pr ∣ 𝑜𝑝 ,
𝛤 (𝜏𝑃 ∣ 𝑜𝑝 ) = 𝜔 (5.19)
( ) ( ) according to the associated utility and the priority of the acquired three
𝑝 𝑝
̂𝐵𝑃 Pr ∣ 𝑜𝑝 + 𝜔
𝛤 (𝜏𝐵 ∣ 𝑜𝑝 ) = 𝜔 ̂𝐵𝑁 Pr ¬ ∣ 𝑜𝑝 , (5.20) regions.
𝑝 ( )
̂𝑁𝑁 Pr ¬ ∣ 𝑜𝑝 .
𝛤 (𝜏𝑁 ∣ 𝑜𝑝 ) = 𝜔 (5.21)
5.5. The algorithm of the proposed model
̂𝑝𝑃 𝑃 , 𝜔
Here, the utility functions 𝜔 ̂𝑝𝐵𝑃 , 𝜔
̂𝑝𝐵𝑁 and 𝜔
̂𝑝𝑁𝑁 are given in Table 3.
Moreover, on the basis of Bayesian procedure, the decision rules for the In this subsection, the steps of the proposed 3WMADM model under
alternative 𝑜𝑝 are illustrated as follows: IVFIIS are summarized by presenting an algorithm to interpret the
method.
(P2) If 𝛤 (𝜏𝑃 ∣ 𝑜𝑝 ) ≥ 𝛤 (𝜏𝐵 ∣ 𝑜𝑝 ) and 𝛤 (𝜏𝑃 ∣ 𝑜𝑝 ) ≥ 𝛤 (𝜏𝑁 ∣ 𝑜𝑝 ), then decide Input: The IVFIIS which describes an IMADM problem and the value
𝑜𝑝 ∈ 𝑃 𝑂𝑆(), of the parameters 𝜃, 𝛿 and 𝜎.
(B2) If 𝛤 (𝜏𝐵 ∣ 𝑜𝑝 ) ≥ 𝛤 (𝜏𝑃 ∣ 𝑜𝑝 ) and 𝛤 (𝜏𝐵 ∣ 𝑜𝑝 ) ≥ 𝛤 (𝜏𝑁 ∣ 𝑜𝑝 ), then decide Output: The alignment and ranking of all alternatives, and an optimal
𝑜𝑝 ∈ 𝐵𝑁𝐷(), alternative.
(N2) If 𝛤 (𝜏𝑁 ∣ 𝑜𝑝 ) ≥ 𝛤 (𝜏𝑃 ∣ 𝑜𝑝 ) and 𝛤 (𝜏𝑁 ∣ 𝑜𝑝 ) ≥ 𝛤 (𝜏𝐵 ∣ 𝑜𝑝 ), then decide Step 1: Transform the IVFIIS into an IVF-CNIIS by converting the
𝑜𝑝 ∈ 𝑁𝐸𝐺(). IVFFN evaluation values into IVF-CN evaluation values by using Def-
inition 3.2.
Ultimately, with the help of Proposition 5.1, the decision rules for the
Step 2: Obtain the possibility dominance classes and the preference
alternative 𝑎𝑝 equivalent to (P2)-(N2) are demonstrated as follows:
degrees (𝐷𝐴 (𝑜𝑝 )) of the alternatives by means of Section 4.
( )
(P2′ ) If Pr ∣ 𝑜𝑝 ≥ 𝛼 ̂𝑝 , then decide 𝑜𝑝 ∈ 𝑃 𝑂𝑆(), Step 3: Acquire the ranking of the alternatives from the pre-decision in
( ) the basis of Section 5.1.
(B2′ ) If 𝛽̂𝑝 < Pr ∣ 𝑜𝑝 < 𝛼 ̂𝑝 , then decide 𝑜𝑝 ∈ 𝐵𝑁𝐷(),
( ) Step 4: Calculate the conditional probabilities of the alternatives
(N2′ ) If Pr ∣ 𝑜𝑝 ≤ 𝛽̂𝑝 , then decide 𝑜𝑝 ∈ 𝑁𝐸𝐺(),
by Eq. (5.15).
̂𝑝 and 𝛽̂𝑝 are calculated as follows:
where the thresholds 𝛼 Step 5: Compute the regret-based relative utility functions of the
alternatives based on the preference degrees obtained in Step 2 and
̂𝑝𝑁𝑁
𝜎𝜔
̂𝑝 =
𝛼 , (5.22) Eqs. (5.17)–(5.18).
̂𝑝𝑁𝑁
𝜎𝜔 ̂𝑝𝑃 𝑃
+ (1 − 𝜎)𝜔 Step 6: Calculate the values of two thresholds 𝛼 ̂𝑝 and 𝛽̂𝑝 by virtue of
̂𝑝𝑁𝑁
(1 − 𝜎)𝜔 Eqs. (5.22)–(5.23).
𝛽̂𝑝 = . (5.23) Step 7: Obtain three regions 𝑃 𝑂𝑆(), 𝐵𝑁𝐷() and 𝑁𝐸𝐺() with the
̂𝑝𝑁𝑁 + 𝜎 𝜔
(1 − 𝜎)𝜔 ̂𝑝𝑃 𝑃
aid of decision rules (P2′ )−(N2′ ).
From Eqs. (5.17)–(5.18) and (5.22)–(5.23), we see that the values of 𝛼
̂𝑝 Step 8: Enumerate the associate utility (𝛬(𝑎𝑝 )) of the alternatives in
and 𝛽̂𝑝 depend on the values of 𝜆𝑝 and 𝜎. light of Eq. (5.24).
Step 9: Rank the alternatives according to the value of associated utility
Remark 5.2. Let 𝜆𝑝 be the preference degree of the alternative 𝑜𝑝 , then in each region and the priority of three regions 𝑃 𝑂𝑆() ≻ 𝐵𝑁𝐷() ≻
we have two special cases: 𝑁𝐸𝐺() (≻ means superiority).
9
A. Mondal et al. Expert Systems With Applications 211 (2023) 118688
Table 4
The transformed IVFIIS for sustainable CBT location selection.
𝑜1 𝑜2 𝑜3 𝑜4
𝑎1 ([0.50, 0.65], [0.60, 0.80]) ([0.45, 0.55], [0.60, 0.75]) ([0.40, 0.50], [0.55, 0.75]) ([0.35, 0.45], [0.65, 0.75])
𝑎2 ∗ ∗ ∗ ([0.30, 0.45], [0.55, 0.70])
𝑎3 ([0.45, 0.50], [0.60, 0.65]) ([0.40, 0.50], [0.70, 0.75]) ([0.45, 0.55], [0.65, 0.75]) ∗
𝑎4 ([0.60, 0.70], [0.45, 0.50]) ([0.65, 0.75], [0.42, 0.60]) ∗ ([0.64, 0.70], [0.35, 0.45])
𝑎5 ∗ ∗ ([0.63, 0.75], [0.44, 0.50]) ([0.60, 0.65], [0.30, 0.40])
𝑎6 ∗ ([0.60, 0.68], [0.50, 0.60]) ([0.64, 0.70], [0.30, 0.35]) ([0.65, 0.75], [0.45, 0.55])
𝑎7 ([0.64, 0.72], [0.38, 0.50]) ([0.70, 0.75], [0.45, 0.57]) ([0.66, 0.74], [0.46, 0.54]) ([0.60, 0.70], [0.48, 0.58])
𝑎8 ([0.63, 0.70], [0.35, 0.50]) ([0.68, 0.76], [0.40, 0.52]) ([0.55, 0.60], [0.42, 0.54]) ∗
Table 5
The transformed IVF-CNIIS for sustainable CBT location selection.
𝑜1 𝑜2 𝑜3 𝑜4
𝑎1 0.5028 + 0.8347𝑖 + 0.6629𝑗 0.4443 + 0.8593𝑖 + 0.6526𝑗 0.4056 + 0.8737𝑖 + 0.6434𝑗 0.3521 + 0.8579𝑖 + 0.6875𝑗
𝑎2 ∗ ∗ ∗ 0.3531 + 0.8948𝑖 + 0.6211𝑗
𝑎3 0.4337 + 0.8878𝑖 + 0.6026𝑗 0.3879 + 0.8412𝑖 + 0.7022𝑗 0.4378 + 0.8493𝑖 + 0.6720𝑗 ∗
𝑎4 0.6294 + 0.8763𝑖 + 0.4270𝑗 0.6677 + 0.8470𝑖 + 0.4555𝑗 ∗ 0.6560 + 0.8755𝑖 + 0.3602𝑗
𝑎5 ∗ ∗ 0.6699 + 0.8571𝑖 + 0.4117𝑗 0.6164 + 0.9010𝑖 + 0.3250𝑗
𝑎6 ∗ 0.6037 + 0.8683𝑖 + 0.5004𝑗 0.6634 + 0.8809𝑖 + 0.2899𝑗 0.6720 + 0.8493𝑖 + 0.4378𝑗
𝑎7 0.6612 + 0.8660𝑖 + 0.3944𝑗 0.6908 + 0.8364𝑖 + 0.4402𝑗 0.6711 + 0.8502𝑖 + 0.4367𝑗 0.6187 + 0.8676𝑖 + 0.4793𝑗
𝑎8 0.6475 + 0.8747𝑖 + 0.3894𝑗 0.6967 + 0.8425𝑗 + 0.3995𝑗 0.5529 + 0.9034𝑖 + 0.4540𝑗 ∗
6. Application Table 6
Possibility dominance class and set of the CBT locations.
In this section, in order to show the validity of the proposed 𝑂 [𝑜𝑝 ]≥𝑎 ≥𝐴 (𝑜𝑝 )
𝑗
3WMADM model, two case studies are presented from Rani and Mishra 𝑜1 {𝑜1 }, 𝑂, {𝑜1 , 𝑜3 , 𝑜4 }, 𝑂, {𝑜1 , 𝑜2 }, 𝑂, {𝑜1 , 𝑜2 , 𝑜3 }, {𝑜1 , 𝑜2 , 𝑜4 } {𝑜1 }
(2022) and Rani et al. (2022), out of which the first one is considered 𝑜2 {𝑜1 , 𝑜2 }, 𝑂, 𝑂, {𝑜2 , 𝑜3 }, 𝑂, 𝑂, {𝑜2 }, {𝑜2 , 𝑜4 } {𝑜2 }
𝑜3 {𝑜1 , 𝑜2 , 𝑜3 }, 𝑂, {𝑜3 , 𝑜4 }, 𝑂, {𝑜1 , 𝑜2 , 𝑜3 }, {𝑜1 , 𝑜3 , 𝑜4 }, {𝑜2 , 𝑜3 }, 𝑂 {𝑜3 }
under IIS and the second one is under complete IS.
𝑜4 𝑂, 𝑂, 𝑂, {𝑜2 , 𝑜3 , 𝑜4 }, 𝑂, {𝑜1 , 𝑜4 }, 𝑂, 𝑂 {𝑜4 }
The first case study is a sustainable CBT location selection problem After that, using Eq. (4.14), the preference degree𝐷𝐴 (𝑜𝑝 ) of each al-
whose description is as follows: ternative is calculated as: 𝐷𝐴 (𝑜1 ) = 0.7577, 𝐷𝐴 (𝑜2 ) = 0.8163, 𝐷𝐴 (𝑜3 ) =
In order to develop sustainable tourism in rural areas of Darjeeling, 0.7103, 𝐷𝐴 (𝑜4 ) = 0.6054.
West Bengal, Indian government wants to select suitable CBT loca-
tion(s) out of four locations: CBTL-I (𝑜1 ), CBTL-II (𝑜2 ), CBTL-III (𝑜3 ) and Step 3: Following Section 5.1, the state set 𝛺 = {, ¬} is obtained,
0.7577 0.8163 0.7103 0.6054
CBTL-IV (𝑜4 ). Based on the literature survey, eight sustainable attributes where = + + + and ¬ is the complement
𝑜1 𝑜2 𝑜3 𝑜4
are identified which are given as 𝐴 = {𝑎1 , 𝑎2 , 𝑎3 , 𝑎4 , 𝑎5 , 𝑎6 , 𝑎7 , 𝑎8 }. The of , in which connotes the fuzzy set of good CBT location and ¬
definitions of the attributes can be found in Rani et al. (2022). The connotes the fuzzy set of bad CBT location. Hence, according to the
weight vector of the attributes is given as 𝑊 = (0.1010, 0.1375, 0.1004, membership degrees, the alternatives are ranked in the pre-decision as
0.1077, 0.1569, 0.1297, 0.1532, 0.1137). Furthermore, the government de- 𝑜2 ≻ 𝑜1 ≻ 𝑜3 ≻ 𝑜4 . Thus, only ranking of the alternatives is obtained,
sires to decide which location(s) should be selected (𝜏𝑃 ), which loca-
but the classification are no achieved in the pre-decision. Therefore, we
tion(s) should not be selected (𝜏𝑁 ), and which location(s) should be
move on to the following steps.
further investigated and take decisions after gathering more informa-
tion (𝜏𝐵 ). In order to meet the requirements of IVFIIS, we randomly Step 4: Using Eq. (5.15), the conditional probability of each alternative
remove IVFFN evaluation values from the IVFFN decision matrix given is estimated and depicted in Table 8.
in Rani et al. (2022), and the resultant IVFIIS is presented in Table 4.
Step 5: By virtue of Eqs. (5.17)–(5.18), the regret-based relative utility
6.1.1. Implementation of the proposed model for case study 1 functions are computed and put in Table 7.
Suppose 𝜃 = 0.5, 𝛿 = 0.3 and 𝜎 = 0.75, then the algorithm, which is Step 6: Two decision thresholds are calculated by means of Eqs. (5.22)–
illustrated in Section 5.5, is implemented in order to solve the presented (5.23), and depicted in Table 8.
problem as follows:
Step 7: From Table 8, the three regions are acquired based on the
Step 1: Using Definition 3.2, the IVFIIS is further transformed into decision rules (P2′ ) − (N2′ ) as: 𝑃 𝑂𝑆() = {𝑜1 , 𝑜2 , 𝑜3 }, 𝐵𝑁𝐷() = {𝑜4 }
an IVF-CNIIS by converting IVFFN evaluation values into the IVF-CN
and 𝑁𝐸𝐺() = 𝜙.
evaluation values. Then, the transformed IVF-CNIIS is illustrated in
Table 5. Step 8: In light of Eq. (5.24), the associated utilities of the alternatives
are calculated as: 𝛬(𝑜1 ) = 0.4627, 𝛬(𝑜2 ) = 0.5176, 𝛬(𝑜3 ) = 0.4102, 𝛬(𝑜4 ) =
Step 2: By means of Section 4, the possibility dominance classes and
0.3311.
possibility dominance set of each alternative are obtained and are
placed in Table 6. Step 9: In fine, the complete ranking of the alternatives is obtained
Therefore, based on Eq. (4.13), the judgement matrix 1 is formu- according to the associated utility and the priority of regions 𝑃 𝑂𝑆() ≻
lated. 𝐵𝑁𝐷() ≻ 𝑁𝐸𝐺() as: 𝑜2 ≻ 𝑜1 ≻ 𝑜3 ≻ 𝑜4 . Thus, based on
⎛ 1 0.6254 0.7464 0.9013⎞ the categorization and ranking result, DM can make the conclusive
⎜ ⎟ decision. Moreover, the final ranking result is fully coherent with the
0.8086 1 0.7699 0.8703⎟
1 = ⎜
⎜0.6285 0.6321 1 0.8703⎟ pre-decision. Moreover, the proposed model effectively discriminate
⎜0.5890 0.6381 0.5890 1 ⎟⎠ the alternatives. It proves the validity of the proposed model.
⎝
10
A. Mondal et al. Expert Systems With Applications 211 (2023) 118688
| 3 |2 | |2
|𝜈 − 𝜈3 | + |𝜈3 − 𝜈3 | +
6.2. Case study 2: E-waste recycling partner selection | 𝐿1 𝐿2 | | 𝑈1 𝑈2 |
)) 1
| 3 |2 | |2 2,
The second case study is an E-waste recycling partner (EWRP) |𝜋 − 𝜋3 | + |𝜋3 − 𝜋3 |
| 𝐿1 𝐿2 | | 𝑈1 𝑈2 |
selection problem whose description is as follows:
𝑠𝑖𝑚(1 , 2 ) = 1 − (1 , 2 ).
For proper management of E-waste, four EWRPs 𝑂 = {𝑜1 , 𝑜2 , 𝑜3 , 𝑜4 }
are chosen by an electronics farm located in Delhi, India for further Therefore, the similarity degree between two IVFFN objects according
assessment with respect to four attributes: recycling performance and to the models of Liu et al. (2016), Yang, Li et al. (2020) and Zhan et al.
delivery history (𝑎1 ), environmental management system (𝑎2 ), reduction (2021) is calculated as follows:
in green house gas emission (𝑎3 ) and recycling cost (𝑎4 ). The weight
⎧
vector of the attributes is 𝑊 = (0.3, 0.25, 0.18, 0.27)𝑇 . Among the ⎪1, 𝑠𝑖𝑚(𝑜𝑝 , 𝑜𝑞 ) ≥ 0.5 ∧ 𝑜𝑝𝑗 ≠∗ ∧𝑜𝑞𝑗 ≠∗,
attributes, only 𝑎4 is cost attribute and others are benefit attributes. ⎪
⎪0, 𝑠𝑖𝑚(𝑜𝑝 , 𝑜𝑞 ) ≤ 0.5 ∧ 𝑜𝑝𝑗 ≠∗ ∧𝑜𝑞𝑗 ≠∗,
Therefore, according to Rani and Mishra (2022), the IVFFN evaluation
⎪ 1
matrix is displayed in Table 9. 𝑆𝐷𝑎𝑗 (𝑜𝑝 , 𝑜𝑞 ) = ⎨ , 𝑜𝑝𝑗 =∗ ∨𝑜𝑞𝑗 =∗,
⎪ ∣ 𝑉𝑎𝑗 ∣
⎪
6.2.1. Implementation of the proposed model for case study 2 ⎪ 1 , 𝑜𝑝𝑗 =∗ ∧𝑜𝑞𝑗 =∗ .
At first, according to Definition 3.2, the IVFFNs are converted ⎪ ∣ 𝑉𝑎𝑗 ∣2
⎩
into IVF-CNs to get the IVF-CN evaluation matrix, which is placed
in Table 10. Thereafter, in the virtue of Section 4, the possibility Furthermore, in the model of Liu et al. (2016), the state set is consid-
dominance class and set of each alternative are obtained and placed in ered as = {𝑜𝑝 ∣ 𝐷𝐴 (𝑜𝑝 ) ≥ 0.8} and the loss functions are replaced
Table 11. Therefore, on the basis of Eq. (4.13), the judgement matrix by the relative utility functions as: ̂𝑙𝑃𝑝 𝑃 = ̂ 𝑝
𝑙𝑁𝑁 = 0, ̂ 𝑝
𝑙𝑁𝑃 = ̂ 𝑢𝑝𝑃 𝑃 , ̂
𝑙𝑃𝑝 𝑁 =
𝑝 𝑝 𝑝 𝑝 𝑝 𝑝 𝑝 𝑝 𝑝 𝑝 𝑝
2 is formulated. 𝑢𝑁𝑁 , 𝑙𝐵𝑃 = 𝜎 𝑙𝑁𝑃 and 𝑙𝐵𝑁 = 𝜎 𝑙𝑃 𝑁 , where 𝑙𝑃 𝑃 , 𝑙𝑃 𝑁 , 𝑙𝐵𝑃 , 𝑙𝐵𝑁 , 𝑙𝑁𝑃 and ̂
̂ ̂ ̂ ̂ ̂ ̂ ̂ ̂ ̂ ̂ 𝑙𝑁𝑁
are the loss functions those are adopted for solving Liu et al.’s model.
⎛ 1 0.70 0.45 0.70⎞
⎜ ⎟ Thus, case study 1 is solved by four models and the obtained results
0.55 1 0.45 0.45⎟
2 = ⎜ are depicted in Table 13. From Table 13, we see that the classification
⎜0.30 0.55 1 0.82⎟
⎜0.30 acquired from the proposed model is almost same as the classification
⎝ 0.55 0.18 1 ⎟⎠
obtained from the three existing models. In all of the models, the CBT
Using Eq. (4.14), the preference degrees of the alternatives are com- locations 𝑜1 and 𝑜2 are always in 𝑃 𝑂𝑆(). Only difference is that 𝑜3 in
puted as: 𝐷𝐴 (𝑜1 ) = 0.6167, 𝐷𝐴 (𝑜2 ) = 0.4667, 𝐷𝐴 (𝑜3 ) = 0.5567, 𝐷𝐴 (𝑜4 ) = 𝑃 𝑂𝑆() of the proposed model and the model of Zhan et al. (2021)
0.3433. Thus, the state set is obtained as: 𝛺 = {, ¬}, where = is shifted to 𝐵𝑁𝐷() in the models of Liu et al. (2016) and Xin et al.
0.6167 0.4667 0.5567 0.3433 (2021). Further, only in the model of Xin et al. (2021), 𝑜4 appears in
+ + + . Then, the ranking of the alternatives
𝑜1 𝑜2 𝑜3 𝑜4 𝑁𝐸𝐺(). Since, in the model of Xin et al. (2021), the decision making
is derived according to the membership degrees as: 𝑜1 ≻ 𝑜3 ≻ 𝑜2 ≻ 𝑜4 .
was taken place after filling the missing data. Therefore, their model
For classification of the alternatives, we step forward to the subsequent
became complete, and in complete IS the dominance classes and sets
steps. In the next step, the conditional probabilities of the alternatives
are estimated and put in Table 12. Lastly, by taking 𝜃 = 0.5, 𝛿 = 0.3 become different from an IIS. Thus, the diversity in the result is logical.
and 𝜎 = 0.65, the regret-based relative utility values are calculated and Moreover, a huge difference in classification result is not seen in Ta-
based on these, the decision thresholds are estimated, which are limned ble 13, therefore we conclude that the proposed model can efficaciously
in Table 12. categorize the alternatives in IMADM problems. Therefore, in view of
From Table 12, the decision regions are achieved based on the the results in Table 13, the locations 𝑜3 and 𝑜4 should be put on hold
decision rules (P2′ ) − (N2′ ) as: 𝑃 𝑂𝑆() = {𝑜1 }, 𝐵𝑁𝐷() = {𝑜2 , 𝑜3 }, while rest of the locations can be chosen as competent locations in case
𝑁𝐸𝐺() = {𝑜4 }. Ultimately, the ranking of the alternatives is obtained study 1.
as follows: 𝑜1 ≻ 𝑜3 ≻ 𝑜2 ≻ 𝑜4 .
7.2. Comparison with the existing models in complete environment
7. Comparative analysis and discussion
In light of the fact that the proposed model can also be utilized to
From Section 6, we see that the proposed model can handle MADM deal with MADM problems in complete environment, this subsection
problems in incomplete as well as complete environment. Therefore, to compares the result of the proposed model with the existing 3WMADM
perfectly unravel the validity and practicability of the proposed model, models in complete environment based on case study 2. For this pur-
in this section, we conduct the comparative analysis with the existing pose, eight 3WMADM models are selected which are the models of Gao
models in different environments. et al. (2020), Huang and Zhan (2021), Jia and Liu (2019, 2021),
11
A. Mondal et al. Expert Systems With Applications 211 (2023) 118688
Table 9
The IVFFN evaluation matrix for EWRP selection.
𝑂 𝑎1 𝑎2 𝑎3 𝑎4
𝑜1 ([0.45, 0.65], [0.55, 0.75]) ([0.60, 0.75], [0.35, 0.50]) ([0.65, 0.75], [0.40, 0.55]) ([0.65, 0.80], [0.40, 0.55])
𝑜2 ([0.65, 0.70], [0.40, 0.75]) ([0.50, 0.60], [0.65, 0.75]) ([0.60, 0.65], [0.50, 0.60]) ([0.65, 0.70], [0.55, 0.65])
𝑜3 ([0.70, 0.80], [0.40, 0.60]) ([0.70, 0.75], [0.30, 0.45]) ([0.55, 0.65], [0.45, 0.55]) ([0.60, 0.65], [0.50, 0.60])
𝑜4 ([0.68, 0.75], [0.45, 0.55]) ([0.65, 0.70], [0.45, 0.60]) ([0.57, 0.65], [0.40, 0.55]) ([0.50, 0.70], [0.50, 0.55])
Table 10
The IVF-CN evaluation matrix for EWRP selection.
𝑂 𝑎1 𝑎2 𝑎3 𝑎4
𝑜1 0.5054 + 0.8575𝑖 + 0.6217𝑗 0.6636 + 0.8665𝑖 + 0.3852𝑗 0.6754 + 0.8512𝑖 + 0.4219𝑗 0.7034 + 0.8350𝑖 + 0.4119𝑗
𝑜2 0.6354 + 0.8557𝑖 + 0.4891𝑗 0.4808 + 0.8434𝑖 + 0.6611𝑗 0.5882 + 0.8741𝑖 + 0.5049𝑗 0.6221 + 0.8466𝑖 + 0.5342𝑗
𝑜3 0.7164 + 0.8204𝑖 + 0.4312𝑗 0.7113 + 0.8452𝑖 + 0.3316𝑗 0.5770 + 0.8911𝑖 + 0.4647𝑗 0.5882 + 0.8741𝑖 + 0.5049𝑗
𝑜4 0.6845 + 0.8425𝑖 + 0.4333𝑗 0.6394 + 0.8584𝑖 + 0.4735𝑗 0.5881 + 0.8912𝑖 + 0.4460𝑗 0.5847 + 0.8830𝑖 + 0.4815𝑗
Table 11 the proposed model over the existing models, which are the main
The possibility dominance class and set for EWRPs. contributions of the proposed model, are demonstrated as follows:
𝑂 [𝑜𝑝 ]≥𝑎 ≥𝐴 (𝑜𝑝 )
𝑗
1. Some existing models (Huang & Zhan, 2021; Jia & Liu, 2019;
𝑜1 𝑂, {𝑜1 , 𝑜2 }, {𝑜1 }, {𝑜1 } {𝑜1 }
𝑜2 {𝑜2 , 𝑜3 , 𝑜4 }, 𝑂, {𝑜1 , 𝑜2 }, {𝑜1 , 𝑜2 } {𝑜2 } Zhu, Ma, and Zhan, 2022) ignored the impreciseness of infor-
𝑜3 {𝑜3 }, {𝑜3 }, 𝑂, {𝑜1 , 𝑜2 , 𝑜3 } {𝑜3 } mation involved in real-life MADM problem. Although some
𝑜4 {𝑜3 , 𝑜4 }, {𝑜1 , 𝑜3 , 𝑜4 }, {𝑜1 , 𝑜2 , 𝑜4 }, 𝑂 {𝑜4 } models (Wang et al., 2022a, 2022c; Xin et al., 2021; Zhan et al.,
2021) considered uncertainty in the MADM, their models either
Table 12 provided very narrow scope or could not handle uncertainty
The conditional probability and the decision thresholds for the EWRPs. efficiently. Thus, their models cannot be exploited to solve the
( )
𝑂 Pr ∣ 𝑜𝑝 𝛼
̂𝑝 𝛽̂𝑝 ̂𝑝𝑃 𝑃
𝜔 𝜔̂𝑝𝐵𝑃 ̂𝑝𝐵𝑁
𝜔 ̂𝑝𝑁𝑁
𝜔 problems under IVFFN environment. On the other hand, the pro-
𝑜1 0.6167 0.5749 0.2816 0.4758 0.3093 0.2252 0.3464 posed model publish the information by IVFFN and consequently
𝑜2 0.4667 0.7414 0.4539 0.3245 0.2110 0.3256 0.5010 by IVF-CN, that can efficiently handle real-world complex and
𝑜3 0.5567 0.6461 0.3461 0.4168 0.2709 0.2664 0.4098 hesitant situation.
𝑜4 0.3433 0.8580 0.6366 0.1902 0.1237 0.4023 0.6189 2. The formulated model can successfully tackle the MADM prob-
lems in both complete and incomplete environments, whereas
the existing 3WMADM models (Gao et al., 2020; Huang & Zhan,
2021; Jia & Liu, 2019; Liang et al., 2018; Wang et al., 2022a,
Liang et al. (2018), Wang et al. (2022a, 2022c) and Zhu, Ma, and
2022c; Zhu, Ma, and Zhan, 2022) developed in complete envi-
Zhan (2022). For fare comparison, all of the models are executed with
ronment can only deal with complete MADM problems. Hence,
IVFFNs. The Euclidean distance between IVFFNs ((1 , 2 )) is utilized
compared to these 3WMADM models, the proposed model has a
in all of the aforementioned models and for the models of Huang and
broader scope of application.
Zhan (2021), Wang et al. (2022a, 2022c) and Zhu, Ma, and Zhan
3. The designed model can take into account the behaviours DM
(2022), the possibility dominance relation is used. Furthermore, the
such as regret and risk in decision making process. However,
conditional probabilities which are derived in the proposed model
there are very few 3WMADM models (Huang & Zhan, 2021; Zhu,
are utilized in the models of Jia and Liu (2019, 2021). Also, the
Ma, and Zhan, 2022) that considered such behaviours. More-
ideal solution in the model of Jia and Liu (2021) is considered as
over, in the model of Zhu, Ma, and Zhan (2022), the regret and
([1, 1], [0, 0]). Thus, case study 2 is solved by different models and the
rejoice are calculated between two objects, this is meaningless
obtained results are showed in Table 14.
for two objects which are in positive region. Also, in their model,
From Table 14, we see that the classification result of the proposed
the results under different strategies can be different, which is
model coincides with the results of most of the models, only a slight
unstable and creates more ambiguity. In contrast, the proposed
differences are noticed. Moreover, the ranking result of the proposed
model reveals actual regret by taking into account regret for
model is slightly different from the ranking result obtained by the
missing the optimal object. Also, the proposed model provides
model of Wang et al. (2022c), but the optimal project is alike in
more certain result. In the model of Huang and Zhan (2021), the
all of the models, i.e., the EWRP 𝑜1 . This substantiates the validity
utilities of the objects are ignored, rather than the objects are
of the proposed model. What is more, the models of Jia and Liu directly utilized in calculating the regret values. That not only
(2019), Liang et al. (2018) and Wang et al. (2022a) cannot provide compromises with the actual performance of the objects, but also
the ranking order of the alternatives. Thereby, the proposed model is ignores risk of decision making. On the other hand, the proposed
fruitful and can derive credible classification and ranking results in model exploits the utilities of the objects by a risk-based utility
complete environment. function. Hence, the proposed model is more advantageous than
the above-mentioned models.
7.3. Discussions 4. The proposed model launches a novel procedure to obtain the
fuzzy state set and accordingly provides a new method to esti-
The comparative analysis exhibits the validity of the proposed mate the conditional probability. This can cut down the draw-
model. In this subsection, the advantages of the proposed model are backs appeared due to the subjectiveness of decision making (Jia
summarized, which can vindicate the superiority of the proposed & Liu, 2019, 2021; Liu et al., 2016; Zhan et al., 2021), or due
model. For this purpose, the limitations of the previous models have to omitting the conditional probability (Yang, Li et al., 2020).
been pointed out via exploring the key differences between the pro- Furthermore, the proposed model also surpasses the limitation
posed model and existing models. Consequently, the elementary dif- of directly calculated conditional probability by distance-based
ferences among different 3WMADM models are put up in Table 15. methods (Gao et al., 2020; Liang et al., 2018). In addition, Wang
Based on the differences mentioned in Table 15, the advantages of et al. (2022c) utilized the ideal-degree to obtain the state set and
12
A. Mondal et al. Expert Systems With Applications 211 (2023) 118688
Table 13
The classification results of different 3WMADM models in incomplete environment.
Model Strategy 𝑃 𝑂𝑆() 𝐵𝑁𝐷() 𝑁𝐸𝐺()
Model in Xin et al. (2021) {𝑜1 , 𝑜2 } {𝑜3 } {𝑜4 }
Optimistic {𝑜1 , 𝑜2 } {𝑜3 , 𝑜4 } 𝜙
Model in Liu et al. (2016)
Pessimistic {𝑜1 , 𝑜2 } {𝑜3 , 𝑜4 } 𝜙
Rule I {𝑜1 , 𝑜2 } {𝑜3 , 𝑜4 } 𝜙
Model in Yang, Li et al. (2020)
Rule II {𝑜1 , 𝑜2 , 𝑜3 } {𝑜4 } 𝜙
Model in Zhan et al. (2021) {𝑜1 , 𝑜2 , 𝑜3 } {𝑜4 } 𝜙
Proposed model {𝑜1 , 𝑜2 , 𝑜3 } {𝑜4 } 𝜙
Table 14
The classification and ranking results of different 3WMADM models in complete environment.
Model Strategy 𝑃 𝑂𝑆() 𝐵𝑁𝐷() 𝑁𝐸𝐺() Ranking result
Model in Liang et al. (2018) {𝑜1 } {𝑜2 , 𝑜3 , 𝑜4 } 𝜙 –
Model in Wang et al. (2022c) {𝑜1 } {𝑜2 , 𝑜3 } {𝑜4 } 𝑜1 ≻ 𝑜2 ≻ 𝑜3 ≻ 𝑜4
Model in Gao et al. (2020) {𝑜1 } {𝑜2 , 𝑜3 , 𝑜4 } 𝜙 𝑜1 ≻ 𝑜3 ≻ 𝑜2 ≻ 𝑜4
Model in Jia and Liu (2019) {𝑜1 } {𝑜2 , 𝑜3 } {𝑜4 } –
Optimistic {𝑜1 } {𝑜2 , 𝑜3 } {𝑜4 } –
Model in Wang et al. (2022a) Neutral {𝑜1 , 𝑜3 } {𝑜2 } {𝑜4 } –
Pessimistic {𝑜1 , 𝑜3 } {𝑜2 } {𝑜4 } –
Model in Zhu, Ma, and Zhan (2022) Optimistic {𝑜1 } {𝑜2 , 𝑜3 } {𝑜4 } 𝑜1 ≻ 𝑜3 ≻ 𝑜2 ≻ 𝑜4
Equable {𝑜1 } {𝑜2 , 𝑜3 } {𝑜4 } 𝑜1 ≻ 𝑜3 ≻ 𝑜2 ≻ 𝑜4
Pessimistic {𝑜1 , 𝑜3 } {𝑜2 } {𝑜4 } 𝑜1 ≻ 𝑜3 ≻ 𝑜2 ≻ 𝑜4
Model in Huang and Zhan (2021) {𝑜1 } {𝑜2 , 𝑜3 } {𝑜4 } 𝑜1 ≻ 𝑜3 ≻ 𝑜2 ≻ 𝑜4
Model in Jia and Liu (2021) {𝑜1 , 𝑜3 } {𝑜2 } {𝑜4 } 𝑜1 ≻ 𝑜3 ≻ 𝑜2 ≻ 𝑜4
Proposed model {𝑜1 } {𝑜2 , 𝑜3 } {𝑜4 } 𝑜1 ≻ 𝑜3 ≻ 𝑜2 ≻ 𝑜4
the conditional probability, which has the same drawbacks as (0.3, 0.2, 0.2, 0.3) is the weight vector of the attributes. The IVFFN eval-
the distance based method. uation value for the MADM is given as:
5. In the 3WMADM models based on loss functions (Liu et al., 2016; ⎛([0.6, 0.6], [0.1, 0.1]) ([0.2, 0.2], [0.6, 0.6]) ([0.6, 0.6], [0.4, 0.4]) ([0.4, 0.4], [0.2, 0.2])⎞
Wang et al., 2022a, 2022c) can cause greater loss due to not ⎜ ⎟
⎜ ([0.3.0.3], [0.2, 0.2]) ([0.2, 0.2], [0.8, 0.8]) ([0.6, 0.6], [0.3, 0.3]) ([0.8, 0.8], [0.1, 0.1])⎟
choosing the alternative with maximum utility value. Besides, ⎜ ⎟
⎜([0.8, 0.8], [0.2, 0.2]) ([0.2, 0.2], [0.6, 0.6]) ([0.6, 0.6], [0.4, 0.4]) ([0.3, 0.3], [0.1, 0.1])⎟
in real-world DM would pick objects with larger utility value, ⎜([0.4, 0.6], [0.1, 0.4])
⎝ ([0.3, 0.4], [0.2, 0.4]) ([0.4, 0.5], [0.6, 0.8]) ([0.3, 0.6], [0.2, 0.5])⎟⎠
other than those with lower loss. Contrariwise, the designed
model with the relative utility functions can prevent such loss Based on the information, the aggregated values of the alternatives
by sorting the alternatives with maximum utility values. More are obtained by using IVFFWAO operator (Rani & Mishra, 2022) as:
than that, in the model of Zhan et al. (2021), the utility function
𝑜𝑊
1
= ([0.5109, 0.5109], [0.2325, 0.2325]),
cannot involve DM’s psychological behaviours. Meanwhile, the
proposed model introduces a new procedure to evaluate the rel- 𝑜𝑊
2
= ([0.6210, 0.6210], [0.2325, 0.2325]),
ative utility functions based on the regret theory that overcomes 𝑜𝑊
3
= ([0.6210, 0.6210], [0.2325, 0.2325]),
the limitations of the above models. 𝑜∗𝑊 = ([0.3575, 0.5536], [0.2024, 0.4913]),
6. In contrast to most of the 3WMADM models (Jia & Liu, 2019,
2021; Liang et al., 2018; Liu et al., 2016; Wang et al., 2022a; and those are obtained by using IVFFWGO operator (Rani & Mishra,
Yang, Li et al., 2020), the proposed model can provide the 2022) as:
classification and ranking of the alternatives, simultaneously.
𝑜𝐺
1
= ([0.4265, 0.4265], [0.3970, 0.3970]),
This can be more supportive for DM to make final decision,
which is a big advantage of the proposed model. 𝑜𝐺
2
= ([0.4265, 0.4265], [0.5202, 0.5202]),
7. Lastly, in light of the demands of decision system, diverse deci- 𝑜𝐺
3
= ([0.4265, 0.4265], [0.3970, 0.3970]),
sion results can be obtained by changing the parameters (𝜃, 𝛿, 𝜎) 𝑜𝐺 = ([0.3464, 0.5335], [0.3723, 0.5797]).
4
of the proposed model based on the decision-making attitude.
This exhibits the adaptability of the proposed model. Further- Then, by using the score functions (Rani & Mishra, 2022), the scores of
more, in compare to most of the models such as Huang and the alternatives are obtained as:
Zhan (2021), Jia and Liu (2021), Wang et al. (2022a) and ℑ(𝑜𝑊 ) = 0.1208, ℑ(𝑜𝑊 ) = 0.2269, ℑ(𝑜𝑊 ) = 0.2269, ℑ(𝑜𝑊 ) = 0.044,
1 2 3 4
Zhu, Ma, and Zhan (2022), etc. the proposed model contains
less number of parameters which alleviates the arbitrariness of and
decision making.
ℑ(𝑜𝐺
1
) = 0.0150, ℑ(𝑜𝐺
2
) = −0.0632, ℑ(𝑜𝐺
3
) = 0.0150, ℑ(𝑜𝐺
4
) = −0.0265.
13
A. Mondal et al. Expert Systems With Applications 211 (2023) 118688
Table 15
The key distinctions among various 3WMADM models.
Models Uncertainty DM’s behaviour Conditional probability Loss/utility functions Ranking State set Parameters
Proposed model IVFFN ✔ Calculated Calculated ✔ Calculated 3
Model in Xin et al. IFN ✘ Calculated Beforehand ✔ ✘ 9
(2021)
Model in Liu et al. Fuzzy ✘ Calculated Beforehand ✘ Beforehand 6𝑚 + 2
(2016)
Model in Yang, Li et al. Fuzzy ✘ ✘ ✘ ✘ Beforehand 2
(2020)
Model in Zhan et al. Fuzzy ✘ Calculated Calculated ✔ Beforehand 2
(2021)
Model in Liang et al. PFN ✘ Calculated Beforehand ✘ ✘ 6
(2018)
Model in Wang et al. IFN ✘ Calculated Calculated ✔ Calculated 2
(2022c)
Model in Gao et al. IFN ✘ Calculated Calculated ✔ ✘ 1
(2020)
Model in Jia and Liu ✘ ✘ Beforehand Calculated ✘ ✘ 𝑚+𝑛
(2019)
Model in Wang et al. hesitant fuzzy ✘ Calculated Calculated ✘ ✘ 𝑛+2
(2022a)
Model in Zhu, Ma, and ✘ ✔ Calculated Calculated ✔ ✘ 4
Zhan (2022)
Model in Huang and ✘ ✔ Calculated Calculated ✔ ✘ 𝑛+2
Zhan (2021)
Model in Jia and Liu IFN ✘ Beforehand Calculated ✔ ✘ 𝑚+𝑛
(2021)
⎛0.5998 + 0.9219𝑖 + 0.0922𝑗 0.1844 + 0.9203𝑖 + 0.5984𝑗 0.5869 + 0.9076𝑖 + 0.3688𝑗 0.3989 + 0.9758𝑖 + 0.1956𝑗 ⎞
⎜ ⎟
⎜0.2992 + 0.9883𝑖 + 0.1982𝑗 0.1575 + 0.7874𝑖 + 0.7979𝑗 0.5946 + 0.9160𝑖 + 0.2766𝑗 0.7997 + 0.7873𝑖 + 0.0787𝑗 ⎟
⎜0.7979 + 0.7874𝑖 + 0.1575𝑗 0.1844 + 0.9203𝑖 + 0.5984𝑗 0.5869 + 0.9076𝑖 + 0.3688𝑗 0.2999 + 0.9906𝑖 + 0.9909𝑗 ⎟
⎜0.5136 + 0.9423𝑖 + 0.3035𝑗 0.3527 + 0.9732𝑖 + 0.3251𝑗 0.3917 + 0.8482𝑖 + 0.6908𝑗 0.4841 + 0.9391𝑖 + 0.3880𝑗 ⎟⎠
⎝
Box I.
𝑜∗3 = 0.4174 + 0.9447𝑖 + 0.4383𝑗, 𝑜∗4 = 0.4433 + 0.9397𝑖 + 0.4366𝑗. is obtained as S3 ≻ S4 ≻ S7 ≻ S9 ≻ S8 ≻ S10 ≻ S2 ≻ S6 ≻ S1 ≻ S5. Thus,
S9 and S8 are selected as optimal suppliers when handling the problem
Thereafter, by using the possibility degree, the possibility degree matrix
with IVFFN, and S3 is selected when IVF-CN is utilized. Let 𝑑𝑖+ and 𝑑𝑖−
is obtained as follows:
be the distances of 𝑖th supplier from the ideal and anti-ideal solutions,
⎛0.5000 0.4880 0.5001 0.5127⎞ 𝑑−
⎜ ⎟ respectively. Then the relative distances, i.e., + 𝑖 − for the suppliers
0.5120 0.5000 0.5125 0.5250⎟ 𝑑𝑖 + 𝑑𝑖
(𝑝𝑠𝑡 )4×4 =⎜ ,
⎜0.4995 0.4875 0.5000 0.5121⎟ S9, S8 and S3 are obtained as 0.536, 0.545 and 0.572, respectively. The
⎜0.4873 0.4750 0.4879 ⎟
0.5000⎠ distances between two IVFFNs and between two IVF-CNs are given as:
⎝ √
1∑
where 𝑝𝑠𝑡 = 𝑝𝑜𝑠(𝑜∗𝑠 ⪰ 𝑜∗𝑡 ). Therefore, the optimal degree 𝜋𝑠 of each 𝑑(1 , 2 ) = (𝜇 3 − 𝜇3 )2 + (𝜈3 − 𝜈3 )2 + (𝜋3 − 𝜋3 )2 and
(∑𝑚 2√ ∙=𝐿,𝑈 1∙ 2∙ 1∙ 2∙ 1∙ 2∙
14
A. Mondal et al. Expert Systems With Applications 211 (2023) 118688
Table 16 Table 17
The comparative results for involving and for avoiding DM’s behaviours. The ranking results for assorted values of 𝜎.
Model 𝑃 𝑂𝑆() 𝐵𝑁𝐷() 𝑁𝐸𝐺() 𝜎 Ranking The optimal EWRP
The proposed model {S3, S4, S7, S9} {S2, S6, S8, S10} {S1, S5} 0.55 𝑜1 ≻ 𝑜3 ≻ 𝑜2 ≻ 𝑜4 𝑜1
Model in Gao et al. {S3, S4, S7, S8, S9} {S2, S10} {S1, S5, S6} 0.60 𝑜1 ≻ 𝑜3 ≻ 𝑜2 ≻ 𝑜4 𝑜1
(2020) 0.65 𝑜1 ≻ 𝑜3 ≻ 𝑜2 ≻ 𝑜4 𝑜1
Model in Zhan et al. {S3, S4, S7, S8, S9, S10} {S2, S6} {S1, S5} 0.70 𝑜1 ≻ 𝑜3 ≻ 𝑜2 ≻ 𝑜4 𝑜1
(2021) 0.75 𝑜1 ≻ 𝑜3 ≻ 𝑜2 ≻ 𝑜4 𝑜1
0.80 𝑜1 ≻ 𝑜3 ≻ 𝑜2 ≻ 𝑜4 𝑜1
0.85 𝑜1 ≻ 𝑜3 ≻ 𝑜2 ≻ 𝑜4 𝑜1
0.90 𝑜1 ≻ 𝑜3 ≻ 𝑜2 ≻ 𝑜4 𝑜1
0.95 𝑜1 ≻ 𝑜3 ≻ 𝑜2 ≻ 𝑜4 𝑜1
other two models. This reflects the behaviours of DM towards decision
1 𝑜1 ≻ 𝑜3 ≻ 𝑜2 ≻ 𝑜4 𝑜1
making. Because of the behavioural parameters (risk and regret aver-
sion) in the utility function, the proposed model tends to divide more
alternatives into the boundary region in order to avoid risk and regret.
Thus, the proposed 3WMADM model with regret-based utility function 8.2. The experiment on the value of 𝜎
is more effective to eliminate risk of decision making than existing cost
function-based 3WMADM models and utility function-based 3WMADM As mentioned in Proposition 5.3, the values of two decision thresh-
models that did not consider any behaviour of DM. Apart from that, olds depend on the value of 𝜎. Accordingly, this subsection investigates
the ranking of the alternatives is obtained as S3 ≻ S7 ≻ S4 ≻ S9 ≻ S10
the effect of the variation of the value of 𝜎 on the classification and
≻ S8 ≻ S6 ≻ S2 ≻ S1 ≻ S5.
ranking results. For fixed 𝜃 = 0.5 and 𝛿 = 0.3, the classification
results for assorted values of 𝜎 are depicted in Fig. 4. Form Fig. 4,
8. Experimental analysis we notice that the number of alternatives in positive and negative
regions decrease, and those in boundary region increase, as the value
In this section, we explore the effect of key parameters i.e., 𝜃, 𝛿 and of 𝜎 increases. Therefore, the value 𝜎 has a certain influence in the
𝜎 to the decision result. For this, we vary 𝜃 from 0.05 to 0.55 with step classification results. However, the ranking results and the optimal
0.1, 𝛿 from 0 to 0.5 with step 0.1 and 𝜎 from 0.5 to 1 with step 0.1. In alternative are coherent for the assorted values of 𝜎, which can be
what follows, based on case study 1, three experiments are performed.
seen in Table 17. Since the ranking of the alternatives depends on the
associated utility value defined in Eq. (5.24), which is proportional to
8.1. The experiments on the values of 𝜃 and 𝛿 the value of 𝜎. Therefore, whatever the value of 𝜎 will be, the ranking
should be invariant. The result of the proposed model also supports
Since the risk aversion coefficient 𝜃 and the regret aversion coef- the result of Zhan et al. (2021). Hence, the proposed model is very
ficient 𝛿 are two elementary components of the regret-based utility stationary in the matter of ranking with respect to 𝜎.
functions. Therefore, the variation of the values of the concerned
parameters directly effects the classification and ranking of the alterna-
tives. Consequently, this subsection explores the impact of the change 9. Conclusions
of the values of 𝜃 and 𝛿 on the proposed model. For fixed 𝛿 = 0.3 and
𝜎 = 0.65, the classification results of EWRPs for different values of 𝜃 are To depict DM’s psychological behaviours in decision making frame-
shown in Fig. 2. Again, for fixed 𝜃 = 0.5 and 𝜎 = 0.65, the classification work, a novel 3WMADM model has been developed with the aid of
results for different values of 𝛿 are also shown in Fig. 2. From Fig. 2, regret theory in this study to deal with MADM problems. The proposed
we see that when the value of 𝜃 increases, the number of alternatives model has been helpful for providing ranking as well as classification
in boundary and negative regions, respectively, increase and decrease. of the alternatives in both incomplete and complete environments. The
It implies that, if DM is risk seeking and wants to make fast decision, validity, superiority and practicability of the proposed model has been
he/she needs to set the value of 𝜃 smaller. On the other hand, if DM is of exhibited via two case studies, comparative and experimental analyses.
opposite nature, the value of 𝜃 needs to be set bigger. An opposite case In a nutshell, the key contributions of the study are summarized as
is noticed for the case of 𝛿. As the value of 𝛿 increases, the number follows:
of alternatives start to shift from boundary to negative region. Thus,
the proposed model is sensitive to the behavioural parameters 𝜃 and 𝛿. • The regret-based 3WMADM model has been described under
However, there is no variation noticed in ranking results for different IVFFN (consequently under IVF-CN) environment, which not only
values of 𝜃 and 𝛿, which are shown in Fig. 3. Therefore, the proposed has reflected DM’s emotion in decision results, but also has illus-
model is very static in terms of ranking with respect to 𝜃 and 𝛿. trated uncertainty of 3WDM via three angles.
15
A. Mondal et al. Expert Systems With Applications 211 (2023) 118688
Acknowledgements
References
16
A. Mondal et al. Expert Systems With Applications 211 (2023) 118688
Liang, D., Wang, M., & Xu, Z. (2019). Heterogeneous multi-attribute nonadditivity Wang, J., Ma, X., Xu, Z., & Zhan, J. (2022b). Regret theory-based three-way decision
fusion for behavioral three-way decisions in interval type-2 fuzzy environment. model in hesitant fuzzy environments and its application to medical decision. IEEE
Information Sciences, 496, 242–263. Transactions on Fuzzy Systems, http://dx.doi.org/10.1109/TFUZZ.2022.3176686.
Liang, D., Xu, Z., Liu, D., & Wu, Y. (2018). Method for three-way decisions using Wang, Y.-M., Yang, J.-B., & Xu, D.-L. (2005). Interval weight generation approaches
ideal TOPSIS solutions at Pythagorean fuzzy information. Information Sciences, 435, based on consistency test and interval comparison matrices. Applied Mathematics
282–295. and Computation, 167(1), 252–273.
Liu, D., Liang, D., & Wang, C. (2016). A novel three-way decision model based on Wang, W., Zhan, J., & Mi, J. (2022c). A three-way decision approach with probabilistic
incomplete information system. Knowledge-Based Systems, 91, 32–45. dominance relations under intuitionistic fuzzy information. Information Sciences,
Liu, J.-B., Malik, M. A., Ayub, N., & Siddiqui, H. M. A. (2020). Distance measures 582, 114–145.
for multiple-attributes decision-making based on connection numbers of set pair Xin, X.-W., Sun, J.-B., Xue, Z.-A., Song, J.-H., & Peng, W.-M. (2021). A novel intuition-
analysis with dual hesitant fuzzy sets. IEEE Access, 8, 9172–9184. istic fuzzy three-way decision model based on an intuitionistic fuzzy incomplete
Liu, P., Wang, Y., Jia, F., & Fujita, H. (2020). A multiple attribute decision mak- information system. International Journal of Machine Learning and Cybernetics, 13,
ing three-way model for intuitionistic fuzzy numbers. International Journal of 907–927.
Approximate Reasoning, 119, 177–203. Xu, Y., Tang, J., & Wang, X. (2020). Three sequential multi-class three-way decision
Loomes, G., & Sugden, R. (1982). Regret theory: An alternative theory of rational choice models. Information Sciences, 537, 62–90.
under uncertainty. The Economic Journal, 92(368), 805–824. Yang, D., Deng, T., & Fujita, H. (2020). Partial-overall dominance three-way decision
Mondal, A., & Roy, S. K. (2021). Multi-objective sustainable opened-and closed- models in interval-valued decision systems. International Journal of Approximate
loop supply chain under mixed uncertainty during COVID-19 pandemic situation. Reasoning, 126, 308–325.
Computers & Industrial Engineering, 159, Article 107453. Yang, X., Li, T., & Tan, A. (2020). Three-way decisions in fuzzy incomplete information
Mondal, A., & Roy, S. K. (2022). Application of choquet integral in interval type-2 systems. International Journal of Machine Learning and Cybernetics, 11(3), 667–674.
Pythagorean fuzzy sustainable supply chain management under risk. International Yao, Y. (2010). Three-way decisions with probabilistic rough sets. Information Sciences,
Journal of Intelligent Systems, 37(1), 217–263. 180(3), 341–353.
Mondal, A., Roy, S. K., & Midya, S. (2021). Intuitionistic fuzzy sustainable multi- Ye, J., Zhan, J., & Sun, B. (2021). A three-way decision method based on fuzzy rough
objective multi-item multi-choice step fixed-charge solid transportation problem. set models under incomplete environments. Information Sciences, 577, 22–48.
Journal of Ambient Intelligence and Humanized Computing, http://dx.doi.org/10.1007/ Yu, C., Shao, Y., Wang, K., & Zhang, L. (2019). A group decision making sustain-
s12652--021--03554--6. able supplier selection approach using extended TOPSIS under interval-valued
Pawlak, Z. (1982). Rough sets. International Journal of Computer & Information Sciences, Pythagorean fuzzy environment. Expert Systems with Applications, 121, 1–17.
11(5), 341–356. Zhan, J., Ye, J., Ding, W., & Liu, P. (2021). A novel three-way decision model based
Rani, P., & Mishra, A. R. (2022). Interval-valued fermatean fuzzy sets with multi-criteria on utility theory in incomplete fuzzy decision systems. IEEE Transactions on Fuzzy
weighted aggregated sum product assessment-based decision analysis framework. Systems, http://dx.doi.org/10.1109/TFUZZ.2021.3078012.
Neural Computing and Applications, 34, 8051–8067. Zhang, H., & Ma, Q. (2020). Three-way decisions with decision-theoretic rough sets
Rani, P., Mishra, A. R., Deveci, M., & Antucheviciene, J. (2022). New complex pro- based on Pythagorean fuzzy covering. Soft Computing, 24(24), 18671–18688.
portional assessment approach using einstein aggregation operators and improved Zhang, Q., Xie, Q., & Wang, G. (2018). A novel three-way decision model with
score function for interval-valued fermatean fuzzy sets. Computers & Industrial decision-theoretic rough sets using utility theory. Knowledge-Based Systems, 159,
Engineering, 169, Article 108165. 321–335.
Roy, S. K., Midya, S., & Weber, G.-W. (2019). Multi-objective multi-item fixed-charge Zhao, K. (1989). Set pair and set pair analysis-a new concept and systematic analysis
solid transportation problem under twofold uncertainty. Neural Computing and method. In Proceedings of the national conference on system theory and regional
Applications, 31(12), 8593–8613. planning (pp. 87–91).
Wang, T., Li, H., Qian, Y., Huang, B., & Zhou, X. (2020). A regret-based three-way Zhu, J., Ma, X., & Zhan, J. (2022). A regret theory-based three-way decision approach
decision model under interval type-2 fuzzy environment. IEEE Transactions on Fuzzy with three strategies. Information Sciences, 595, 89–118.
Systems, 30(1), 175–189. Zhu, J., Ma, X., Zhan, J., & Yao, Y. (2022). A three-way multi-attribute decision
Wang, T., Li, H., Zhang, L., Zhou, X., & Huang, B. (2020). A three-way decision model making method based on regret theory and its application to medical data in fuzzy
based on cumulative prospect theory. Information Sciences, 519, 74–92. environments. Applied Soft Computing, 123, http://dx.doi.org/10.1016/j.asoc.2022.
Wang, J., Ma, X., Xu, Z., & Zhan, J. (2022a). A three-way decision approach with risk 108975.
strategies in hesitant fuzzy decision information systems. Information Sciences, 588,
293–314.
17