Author’s Accepted Manuscript

Supplier selection and order allocation with green

criteria: An MCDM and multi-objective
optimization approach

Sadeque Hamdan, Ali Cheaitou

www.elsevier.com/locate/caor

PII: S0305-0548(16)30269-6
DOI: http://dx.doi.org/10.1016/j.cor.2016.11.005
Reference: CAOR4118
To appear in: Computers and Operation Research
Received date: 20 July 2016
Accepted date: 7 November 2016
Cite this article as: Sadeque Hamdan and Ali Cheaitou, Supplier selection and
order allocation with green criteria: An MCDM and multi-objective optimization
a p p r o a c h , Computers and Operation Research,
http://dx.doi.org/10.1016/j.cor.2016.11.005
This is a PDF file of an unedited manuscript that has been accepted for
publication. As a service to our customers we are providing this early version of
the manuscript. The manuscript will undergo copyediting, typesetting, and
review of the resulting galley proof before it is published in its final citable form.
Please note that during the production process errors may be discovered which
could affect the content, and all legal disclaimers that apply to the journal pertain.
Supplier selection and order allocation with green criteria: An MCDM and
multi-objective optimization approach
Sadeque HAMDANa and Ali CHEAITOUa*
a
SEAM Research group and Industrial Engineering and Engineering Management Department,
College of Engineering, University of Sharjah, PO. Box 27272, Sharjah, United Arab Emirates
[email protected]

[email protected]

[email protected]
*
Corresponding author. Tel.: +971(0)65053921; fax: +971(0)65053963.
Abstract
This research provides a decision-making tool to solve a multi-period green supplier selection and order
allocation problem. The tool contains three integrated components. First, fuzzy TOPSIS (technique for order of
preference by similarity to ideal solution) is used to assign two preference weights to each potential supplier
according to two sets of criteria taken separately: traditional and green. Second, top management uses an
analytic hierarchy process (AHP) to assign a global importance weight to each of the two sets of criteria based
on the strategy of the company and independently of the potential suppliers. Third, for each supplier, the
preference weight obtained from fuzzy TOPSIS regarding the traditional criteria is then multiplied by the global
importance weight of the set of traditional criteria. The same is done for the green criteria. The two combined
preference weights obtained for each supplier are then used in addition to total cost to select the best suppliers
and to allocate orders using multi-period bi-objective and multi-objective optimization. The mathematical
models are solved using the weighted comprehensive criterion method and the branch-and-cut algorithm. The
approach of this research has a major advantage: it provides top management with flexibility in giving more or
less importance weight to green or traditional criteria regardless of the number of criteria in each category
through the use of AHP, which reduces the effect of the number of criteria on the preference weight of the
suppliers. Contrary to the case in which each supplier is evaluated on the basis of all criteria at the same time,
the proposed approach would not necessarily result in a supplier with poor green performance being ranked
among the best for a situation in which the number of green criteria is smaller than the number of traditional
criteria. In this case, the final ranking would mainly depend on the global weight of the green criteria set given
by the top management using AHP as well as on the ranking of the supplier in terms of green criteria obtained
from fuzzy TOPSIS. Extensive numerical experiments are conducted in which the bi-objective and multi-
objective models are compared and the effect of the separation between green and traditional criteria is
investigated. The results show that the two optimization approaches provide very close solutions, which leads to
a preference for the bi-objective approach because of its lower computation time. Moreover, the results confirm
the flexibility of the proposed approach and show that combining all criteria in one set is a special case. Finally,
a time study is performed, which shows that the bi-objective integer linear programming model has a
polynomial computation time.

Keywords: Supplier selection; Planning; Order allocation; Green criteria; Multi-criteria decision making;
Optimization.

1. Introduction
Supply chain management aims to maximize a firm’s profitability and customer satisfaction through integrated
planning and control decisions [1]. Moreover, globalization has increased the challenges of supply chains to
remain competitive by focusing on reducing operational costs and enlarging overall profit [2–4]. However,
because of the increased awareness of environmental issues, both public and private sectors face increasing

1
pressure to consider environmental aspects in managing their operations and supply chains. This consideration is
recognized as green supply chain management [5–7].

Green supply chain management is the management of the flows of funds, information, and products between
and among all stages of the supply chain to find the right balance between environmental and economic aspects
[5,8]. Moreover, purchasing is an essential strategic function in supply chain management for the supply chain
to be competitive. Purchasing has six major decision processes: making or buying, supplier selection, contract
negotiation, design collaboration, procurement, and sourcing analysis [1,9,10]. Among these six processes,
supplier selection has received considerable attention since the 1960s to achieve an effective purchasing strategy
because the right selection reduces cost and improves quality of final products [1,2,11].

Supplier selection in which raw materials and components represent a significant percentage of the total product
cost is the most critical issue firms face [9]. The importance of supplier selection has increased also from
outsourcing initiatives in which firms rely more on suppliers to improve the quality of their products, to reduce
their costs, or to focus on a specific part of their operations [12]; thus, supplier selection constitutes a strategic
decision [13].

The supplier selection problem can be either a single-sourcing problem, in which one supplier is selected to
satisfy the firm’s entire demand, or a multiple-sourcing problem, in which more than one supplier is selected. In
the latter case, each of the suppliers provides a certain amount of the product based on a set of constraints [13].
Relying on one supplier only increases disruption risk in the supply chain, while depending on multiple sources
incurs an increase in the fixed ordering cost, in terms of administrative and negotiation costs [14]. However,
multiple sourcing is preferred to single sourcing because it allows for order flexibility [13].

Supplier selection is a complex, multi-criteria decision-making (MCDM) process that considers both qualitative
and quantitative factors to choose reliable suppliers [2,13,15,16]. This complexity comes from unpredictable and
uncontrollable factors that may be uncertain and conflict with each other [16,17]. In the past, supplier selection
and order allocation problems did not consider environmental effects and issues and focused instead on product
cost, delivery time, and quality [5,6], which we consider in this article as part of the traditional criteria.

Recently, as a result of governmental regulations, increased awareness to protect the environment, and growing
environmental concerns, the trend is to select green vendors because a company’s environmental performance is
measured by both its inter-environmental practices and the environmental image of its suppliers [8,13]. Green
supplier selection is the evaluation of the suppliers to be selected, based on a set of green criteria that includes
environmental aspects, such as waste reduction, use of recyclable materials, green knowledge transfer, and eco-
design. Such green criteria used in supplier selection models are mainly based on company requirements and the
best industrial practices [18]. When evaluating and selecting green suppliers, each green criterion will be given
an importance weight, and each alternative supplier will be evaluated and ranked with respect to these criteria
[19].

In general, planning in the supplier selection problem is a process in which the supplier selection problem is
considered under time-varying demand, cost, and capacity, with the aim to minimize the company’s total cost.
Supplier selection and order allocation problems with planning decisions are called “dynamic supplier selection
problems” as they take into consideration the dynamic environment that might vary in any period of the
planning horizon. The fixed cost of suppliers, demand of the product, and green assessments can change from
one period to another. This change might yield a different selection and planning schedule, in which a supplier
may be used in one period but not in the other periods. The supplier selection and order allocation problem
might also serve as a re-assessment problem in which the suppliers are re-evaluated with new order allocation.

The purpose of this article is to develop a decision-making tool based on MCDM and bi-objective and multi-
objective optimization to help decision makers integrate green criteria in the finite-horizon multi-period supplier
selection and order allocation problem. The proposed tool deals with deterministic demand and the multiple-
sourcing situation. The tool is structured in three steps: first, it gives to every potential supplier two preference
2
weights based on traditional and green criteria taken separately using fuzzy TOPSIS (technique for order of
preference by similarity to ideal solution). Second, it assigns an importance weight to the set of traditional
criteria and another importance weight to the set of green criteria using the analytic hierarchy process (AHP).
This step is usually taken by top management in accordance with the company's strategy and independent of the
available suppliers and the number of criteria in each category. Third, for each supplier, the preference weight
obtained from fuzzy TOPSIS regarding the traditional criteria is multiplied by the global importance weight of
the set of traditional criteria. The same is done for the green criteria and for every supplier. The two combined
preference weights obtained for each supplier (traditional and green), in addition to the fixed and variable
purchasing costs and the inventory holding and penalty shortage costs, serve as input for multi-period bi- and
multi-objective optimization models, which are solved with the weighted comprehensive criterion method
(WCCM) and the branch-and-cut algorithm using the Optimization Toolbox of MATLAB.

We can summarize the major advantage of this approach as follows: if, for example, the company is developing
or following a strategy that focuses on the green aspect of the business, top management will assign a higher
importance weight to the set of green criteria. Therefore, a supplier that has low traditional performance, such as
a high cost, but excellent green performance will rank among the best suppliers and will be eventually selected,
even if the number of green criteria considered is lower than the number of traditional criteria. Moreover, the
methodological choice is justified for three reasons. First, the use of fuzzy TOPSIS does not create any
inconsistency issues [20]. At the same time, fuzzy TOPSIS allows the model to account for any uncertainty that
may come from the experts while ranking the different alternatives on the basis of the traditional or green
criteria, especially when the number of alternatives or criteria is large [20]. The use of AHP in the second step
makes it simple for top management and does not lead to any inconsistency because the number of alternatives
is equal to 2 (green criteria set and traditional criteria set), which we discuss in section 3.2.1. Third, the
optimization models will provide the organization with the optimal suppliers to deal with in each period, based
on the suppliers' evaluation (fuzzy TOPSIS), top management choices (AHP), the suppliers' relative costs and
availability, and the company's inventory holding and shortage costs.

We implemented this approach in software we developed using MATLAB. The software has a user-friendly
graphical user interface (GUI) in which users from the organization's top management can make their pairwise
comparison easily by entering the relative weights of the two groups of criteria. Moreover, the other decision
makers in charge of the suppliers' evaluation can also enter the number of criteria on which they want to
evaluate the suppliers, the number of suppliers (alternatives), the weight of each criterion in every set, and the
relative weight of each supplier with respect to every criterion using linguistic variables. The cost elements,
demand, and suppliers’ capacities can also be entered easily using an MS Excel file imported by the software.
The software processes the fuzzy TOPSIS and AHP calculations, feeds in an optimization code based on the bi-
objective model, and uses then the Optimization Toolbox of MATLAB to solve it. The supply chain
departments of two major companies in the Gulf Cooperation Council region have used this software on two
case studies: a major aluminum manufacturing company and a major facilities management company. The
purpose was to validate the approach and the software. Section 5 presents one of the case studies.

The rest of this article proceeds as follows: section 2 presents the literature review and the research findings, and
section 3 defines the proposed model and the adopted solution approach. Section 4 exhibits the numerical
experiments, and section 5 concludes with a case study. Section 6 is dedicated to general conclusion.

2. Literature review
We classified the literature related to our work into four categories. The first category includes the works on
supplier selection and order allocation. The second category evaluates the studies with a focus on green criteria.
The third category explores the literature on green supplier selection and green supplier selection with order
allocation. Finally, the fourth category includes the works on supplier selection and planning problems. We then
summarize the main findings from the literature and discuss our research’s contribution.

3
2.1 Supplier selection and order allocation

Supplier selection is an activity recognized as the most important and prominent part of the purchasing function
as it contributes to enhancing competitive strategy and global market share by reducing operational costs (e.g.,
maintenance cost), offering high-quality products, enlarging total supply chain profit, and improving total
supply chain performance [14,16,21,22]. In today’s global competition, customer expectations have increased;
these expectations cannot be explained only by lower product cost, as they also involve quality, lead time,
warranties, and many other criteria that make supplier selection a multi-criteria problem. For supplier selection
problems, Dickson [23] proposed 23 different criteria used as reference in the majority of supplier selection
studies [16].

In the literature, two types of supplier selection problems exist. The first type is the single-sourcing supplier
selection problem, in which a single supplier can fully satisfy the demands of the company and the company
only needs to decide which supplier to select. Chou and Chang [24] used a strategy-aligned fuzzy simple multi-
attribute rating technique (SMART) for solving the vendor selection problem. Wang et al. [25] proposed the
fuzzy hierarchical TOPSIS methodology for supplier selection problems to provide more accurate weights.
Kwong et al. [26] developed an integrated two-stage model based on fuzzy set theory and SMART to assess the
performance of suppliers. Moreover, Bruno et al. [27] used AHP combined with fuzzy set theory, Tavana et al.
[28] applied artificial neural network, Rahiminezhad et al. [29] used a balanced scorecard with fuzzy AHP, and
Memon et al. [30] implemented grey system theory with uncertainty theory in the supplier selection problem.
Shemshadi et al. [31] applied fuzzy VIKOR (VIseKriterijumska Optimizacija I Kompromisno Resenje) based on
an entropy measure for objective weighting in the supplier selection problem. They extended the VIKOR
method with a mechanism to extract and deploy objective weights on the basis of the Shannon entropy concept.

The second type is the multiple-sourcing supplier selection problem in which a single supplier cannot satisfy all
the demand, and thus there is a need to split the orders among other suppliers. In this type, the company must
make two decisions: which suppliers to select and what amount to order from each selected supplier [15,17].

To balance the tradeoff between these conflicting criteria, researchers have developed many models, ranging
from single approaches to hybrid approaches. In a single approach, researchers usually focus on solving the
supplier selection problem and determining the optimum quantities subjected to certain constraints. For
example, Zhang and Zhang [14] developed a mixed-integer programming (MIP) model to minimize the total
cost, including the product cost and fixed cost, with stochastic demand, while assuming that all available
suppliers met the qualitative criteria level. Du et al. [32] investigated the supplier selection problem while taking
into consideration life-cycle cost using a bi-objective model that accounts for operational cost in addition to the
purchasing cost, as minimizing purchasing cost might only lead to more equipment failures, resulting in
increased maintenance cost. They proposed a hybridization of the Pareto genetic algorithm and multi-
intersection and similarity crossover strategy to solve the model. Karpak et al. [33] used goal programming (GP)
to minimize the total product cost and maximize the total product quality and delivery reliability in selecting
suppliers for the hydraulic pump division of a U.S. original equipment manufacturer. Karpak et al. [34] applied
visual interactive GP in single- and multiple-product supplier selection problems. Karpak et al. [35] developed a
GP model to select the suppliers and allocate the quantities on the basis of the total product cost, the total
quality, and delivery reliability. The model is constrained by demand and capacity. Moreover, Kumar et al. [36]
formulated and used fuzzy mixed-integer GP to solve the vendor selection problem with a fuzzy nature. Amorim
et al. [37] proposed an MIP model for stochastic supplier selection in the food industry. Moghaddam [38]
applied Monte Carlo simulation with fuzzy GP to solve the supplier selection problem.

Conversely, for hybrid models, Amin et al. [39] were the first to consider strategic perspectives by developing a
two-stage integrated quantified SWOT analysis technique with fuzzy linear programming (fuzzy LP) to deal
with the supplier selection problem. Some researchers have tried to use historical data in supplier selection, such
as Faez et al. [11], who proposed an integrated case-based reasoning with MIP to select the suppliers and the
quantities to be ordered. Xia and Wu [40] proposed an integrated approach of AHP, improved by rough set

4
theory and multi-objectives mixed-integer linear programming (MILP), for a multi-product supplier selection
and order allocation problem, in which suppliers offer price discounts on total business volumes.

Moreover, Demirtas and Üstün [41] developed a two-stage supplier selection and order allocation model, based
on the analytical network process (ANP) and the multi-objective mixed-integer programming (MOMIP), that
maximizes the purchasing value and minimizes the budget and the defect rates. The model was solved by the ϵ -
constraint method and reservation level using the Tchebycheff procedure. Wu et al. [22] used the Delphi
method, ANP, and the MOMIP model for the supplier selection problem, in which criteria are generated by
experts using the Delphi method, then these criteria serve as input for ANP, and finally, the MOMIP model is
used to select the best suppliers and the associated quantities.

Furthermore, Lee et al. [42] used fuzzy AHP and fuzzy multiple goal programming to select the suppliers for the
thin-film-transistor liquid-crystal display. Liao and Kao [43] developed a two-stage model that uses fuzzy
TOPSIS and multi-choice GP for supplier selection and order allocation in watch manufacturing. Rouyendegh
and Saputro [44] applied fuzzy TOPSIS and multi-choice GP in a fertilizer and chemicals company. Kilic [17]
applied fuzzy TOPSIS with MILP to select the best suppliers for multi-item in a multi-supplier problem, and
Perçin [45] used an integrated AHP–GP approach for supplier selection. Ghorbani et al. [46] used SWOT
analysis and entropy to evaluate suppliers and the integer linear programming (ILP) model to select and
determine the quantities. Sodenkamp et al. [47] used group decision making with different voting power and
linear programming (LP) in supplier selection.

2.2 Green criteria

In today's open market, suppliers that provide environmentally friendly products and services gain more
recognition and respect for their efforts [7]. Literature reviews on green supplier selection are limited, and thus
this field is immature [7,12]. The green idea differs from the sustainability concept, in that sustainability covers
economic, social, and environmental criteria, while green features are limited to environmental criteria. As a
result, research recognizes the green concept as part of the sustainability concept [7].

Govindan et al. [12] and Igarashi et al. [7] reviewed green supplier selection articles. Igarashi et al. [7] found
that these works has used both environmental qualitative and quantitative criteria in green supplier selection
models without clearly identifying the criteria characteristics or even the criterion definition. Green criteria can
be structured into two groups: product-oriented and organization-oriented criteria. Igarashi et al. [7] classified
the main green criteria used in the literature, as shown in Table 1. They also noted that some of the reviewed
articles used either product- or organization-related criteria. Govindan et al. [12] found that the most widely
used approach in green supplier selection is the fuzzy-based single approach, in which the most common green
criterion is the environmental management system, followed by green image, environmental performance, and
environmental competencies.

Table 1 Green criteria classification.

Product-related Organization-related
Use of toxic substances Certification of environmental management system
Use of resources Environmental policy
Environmental labeling Compliance to regulation
Recycled packaging Evaluation of second tier suppliers
Green technology Staff training to increase awareness about environmental issues
Green market share

2.3 Green supplier selection and green supplier selection with order allocation

With environmental regulations and the awareness to protect the environment, there is a need to adopt
environmental aspects in selecting any organization’s suppliers. A green supplier selection problem can be
defined as a classic supplier selection problem, in which environmental criteria are considered along with other
5
traditional criteria in selecting the company’s suppliers [48]. Similar to the classic supplier selection problem, a
green supplier selection problem can be either a single-sourcing or a multiple-sourcing problem. To solve the
green supplier selection problem, researchers have applied different methods.

For example, Kannan et al. [18] proposed a framework for selecting green suppliers using fuzzy TOPSIS.
Büyüközkan and Çifçi [8] developed a hybrid model that used fuzzy DEMATEL (Decision-Making Trial and
Evaluation Laboratory) to extract the mutual relationships of interdependencies with criteria and strength of
interdependence, fuzzy ANP to handle the relationships between the factors, and, finally, fuzzy TOPSIS to
select the best alternative. Amindoust et al. [19] applied fuzzy logic and a new ranking model based on the fuzzy
inference system. Kannan et al. [49] proposed a fuzzy axiomatic design, and Hsu et al. [50] used DEMATEL to
identify the key green criterion in supplier selection problems. Moreover, Freeman and Chen [51] used an
integrated AHP, entropy, and TOPSIS to evaluate the ranking of suppliers while considering green aspects.
Darabi and Heydari [52] used the interval-valued hesitant fuzzy set method. Galankashi et al. [53] applied fuzzy
ANP, Hashemi et al. [54] used ANP and improved grey relational analysis, and Dobos and Vörösmarty [55]
applied data envelopment analysis–type composite indicators with common weight analysis.

However, the previously described methods rank the suppliers without considering capacity limitations or
specifying the order quantities from each supplier. Yeh and Chuang [5] aimed to minimize the total product cost
and the total time while maximizing quality and green score in multiple supply chain stages for a multi-product
green supplier selection problem by applying multi-objective genetic algorithms. They used two algorithms to
find the Pareto-optimal solution set. However, the proposed model was constrained by production capacity
limitation, demand satisfaction for both warehouses and end customers, and some other constraints for criterion
score value.

Moreover, Shaw et al. [6] proposed a fuzzy AHP and fuzzy multi-objective linear programing (MOLP) model to
solve the green supplier selection problem and deal with its vagueness due to the subjectivity in human
decisions. Kannan et al. [13] proposed a hybrid model of fuzzy AHP, fuzzy TOPSIS, and the MOLP approach
to solve multi-sourcing green supplier selection problems. They converted the proposed MOLP model into a
single objective function model by applying a max-min formulation. Zouggari and Benyoucef [56] proposed a
four-stage simulation-based knowledge approach for green supplier selection and order allocation problem

2.4 Supplier selection and order allocation with planning

Most studies in the literature have focused on single-period supplier selection and order allocation problems. As
the selected suppliers for the one-period problem might not be the best choice for other periods in the future,
especially in a dynamic environment, it is important to consider multi-period problems. In a multi-period
configuration, the most suitable suppliers are selected for each period to satisfy the constraints [10], which
makes the supplier selection problem with order allocation a planning problem.

Li et al. [57] proposed a two-stage mathematical model to deal with a material supplier selection and order
allocation problem over a planning horizon in a short product-life-cycle environment. They applied a fuzzy
extended AHP to generate risk weights for different suppliers among five criteria: cost, quality, risk, profile, and
service performance. In the second stage, they developed a bi-objective mathematical model to minimize risk
and cost. They solved this bi-objective model using a dynamic programming algorithm to obtain the supplier
ranks and the orders allocation. The model assumed that each supplier could provide all materials with different
capacities and no inventory.

Lee et al. [58] constructed an integrated lot-sizing model with multi-period supplier selection. This integrated
model minimized the cost under both all-unit and incremental quantity discounts. The model assumed that no
shortage is allowed and demand is known. To solve the proposed model, they adopted two approaches based on
the size of the problem: MIP is used for small-scale problems, while a genetic algorithm (GA) is applied for
complex and large-scale problems.

6
Ware et al. [10] developed mixed-integer non-linear programming to minimize the total cost of purchasing
(TCP) in a dynamic supplier selection problem. The developed model considered a multi-product, multi-period
situation, in which purchasing cost, total transportation cost, penalty cost for not meeting quality requirements,
and delay cost are considered in the objective function. The model was constrained by capacity limitation,
demand, and minimum quality level, and it can be solved using LINGO software.

All the previously stated models evaluate the supplier selection problem without considering environmental
criteria. The only work in which green supplier selection and planning were considered was that of Mafakheri et
al. [1]. They introduced a two-stage dynamic programming model for a single-product multi-period green
vendor selection and order allocation problem. Using AHP in the first stage, they ranked potential suppliers
using four main criteria: price performance, delivery performance, environmental performance, and quality. The
model objectives were to maximize a utility function and to minimize the TCP (including purchasing cost and
inventory holding cost), which they later combined into one objective function and solved using a dynamic
programming algorithm. However, the model is constrained by capacity limitations, deterministic demand, and
inventory allowance.

2.5 Literature review findings and article contribution

The literature on supplier selection is extensive. We reviewed the most relevant studies to our work and
classified them on the basis of the used approach, the used method, and whether they include green aspects, as
well as on some additional aspects. Table 2 lists the classification

Table 2: Literature classification.

Green Additional
Approaches Article Method
aspects comments
Chou and Chang [24] Fuzzy SMART No
Wang et al. [25] Fuzzy hierarchical TOPSIS No
Ranking and
Kwong et al. [26] Fuzzy set theory + SMART No
assessment only
Bruno et al. [27] AHP + fuzzy set theory No
Artificial neural network + Adaptive
Tavana et al.[28] No
Neuro Fuzzy Inference System
Rahiminezhad et al. [29] Balanced score card + fuzzy AHP No
Grey system theory + uncertainty
Memon et al. [30] No
theory
Shemshadi et al. [31] Fuzzy VIKOR + Shannon entropy No
Kannan et al. [18] Fuzzy TOPSIS Yes
fuzzy DEMATEL + fuzzy ANP +
MCDM only Büyüközkan and Çifçi [8] Yes
fuzzy TOPSIS
Amindoust et al. [19] Fuzzy inference system Yes
Kannan et al. [49] Fuzzy axiomatic design Yes
To identify key
Hsu et al. [50] DEMATEL Yes green criteria in
supplier selection
Freeman and Chen [51] AHP + entropy + TOPSIS Yes
Darabi and Heydari [52] Interval-valued hesitant fuzzy set Yes
Galankashi et al. [53] Fuzzy Analytical Network Process Yes
ANP + improved grey relational
Hashemi et al. [54] Yes
analysis
Data envelopment analysis–type
Dobos and Vörösmarty
composite indicators with common Yes
[55]
weight analysis
Optimization Zhang and Zhang [14] MIP No

7
 only Considered life-
cycle cost, solved
using Pareto GA
Du et al [32] GA No and multi-
intersection and
similarity
crossover strategy
Karpak et al. [33] GP No
Single + multiple
Karpak et al. [34] Visual interactive goal No
products
Karpak et al. [35] GP No
Handles realistic
Kumar et al. [36] Fuzzy mixed-integer GP No situation in a fuzzy
environment
Stochastic model
for supplier
Amorim et al. [37] MIP No
selection in food
industry
Monte Carlo simulation with fuzzy
Moghaddam [38] No
GP
With planning and
all-unit quantity
Lee et al. [58] MIP No discount and
incremental
discounts
Mixed-integer non-linear With planning for
Ware et al. [10] No
programming multi-items
Amin et al. [39] SWOT + fuzzy LP No
Used historical
Faez et al. [11] Case-based reasoning + MIP No
data
Price discounts on
AHP + rough set theory + multi-
Xia and Wu [40] No total business
objectives MILP
volumes
Solved using ϵ -
Demirtas and Üstün [41] ANP + MOMIP No
constraint
Wu et al. [22] Delphi + ANP + MOMIP No
Lee et al. [42] Fuzzy AHP + fuzzy multiple GP No
In watch
Liao and Kao [43] Fuzzy TOPSIS + multi-choice GP No
manufacturing
In fertilizer and
Rouyendegh and Saputro
Fuzzy TOPSIS + multi-choice GP No chemicals
[44]
company
Hybrid Multi-item
Kilic [17] Fuzzy TOPSIS + MILP No
Approach problem
(Optimization
+ MCDM) Perçin [45] AHP + GP No

Ghorbani et al. [46] SWOT + Shannon entropy + ILP No

Group decision making with different

Sodenkamp et al. [47] No
voting power + LP
Yeh and Chuang [5] Multi-objective model using GA Yes
Shaw et al. [6] Fuzzy AHP + fuzzy MOLP Yes
Fuzzy AHP + fuzzy TOPSIS + fuzzy Solved using max-
Kannan et al. [13] Yes
MOLP min formulation
Zouggari and Benyoucef Simulation-based knowledge
Yes
[56] approach
Fuzzy extended AHP + bi-objective
Li et al. [57] No With planning
model
With planning for
Mafakheri et al. [1] AHP + dynamic programming Yes
single product

8
We can summarize the findings from the literature review as follows:

1. In the literature, most of the existing models for the supplier selection problem did not account for a
multi-period situation and did not take green criteria into consideration.
2. Some studies select the best green supplier but without considering order allocation or the multi-period
framework.
3. Most of the models in the literature on supplier selection and order allocation with planning did not
account for shortage and did not consider the green aspects.
4. The only reviewed article on green supplier selection with planning [1] did not allow for shortage and
did not consider the fixed ordering cost.
5. To the best of our knowledge, none of the reviewed articles assessed the effect of a two-step ranking
process for suppliers using fuzzy TOPSIS and AHP, thus resulting in two separate weights for each
supplier, while providing the company's top management with flexibility regarding the importance
weights of the green criteria and traditional criteria.

The contributions of the current research are fourfold. First, it integrates green aspects into the supplier
evaluation process, while separating green criteria from traditional criteria in the suppliers ranking process; that
is, we perform the suppliers’ evaluation under the traditional criteria and then under the green criteria using
fuzzy TOPSIS. After that, we carry out a pairwise comparison between the traditional criteria as a group on the
one hand and the green criteria as another group on the other hand, which gives the decision maker the ability to
assign different importance weights to each category of criteria using AHP. We combined the results from the
two methods together in a final importance weight for each supplier. This approach provides the decision maker
with flexibility in evaluating the available suppliers. The use of AHP is justified by its simplicity and by the
absence of inconsistency when the number of alternatives is equal to two (green criteria set and traditional
criteria set), as shown in section 3.2.1. The second contribution is related to the optimization models. To the best
of our knowledge, the only article on green supplier selection and order allocation with multi-period planning
horizon is that of Mafakheri et al. [1], who use AHP to rank all the criteria in a single step. Their approach
makes the model inconsistent when the number of criteria or alternatives changes [20]. In our work, we use
fuzzy TOPSIS to overcome this issue because fuzzy TOPSIS does not have this inconsistency issue [20].
Moreover, this technique allows the model to account for any uncertainty that may arise from the experts while
ranking the different alternatives on the basis of the traditional or green criteria, especially when the number of
alternatives or criteria is large [20]. In addition, the optimization models account for fixed costs and allow the
decision maker to use shortage, which is not the case in the model proposed by Mafakheri et al. [1]. The third
contribution is also related to the optimization models. To the best of our knowledge, this work is the first in the
literature to compare the optimal solutions obtained from two supplier selection and order allocation models.
The first model is a bi-objective optimization model in which we combine the two weights of the suppliers
based on both categories of criteria (green and traditional) into a single objective function, while the second
objective function represents the total cost. The second model is a multi-objective optimization model that
considers the suppliers combined weights based on green criteria in one objective function, while the combined
weights based on traditional criteria are considered in a separate objective function. The third objective function
is the total cost. As the fourth contribution, this work provides a time study for the bi-objective optimization
model.

3. Models
As explained previously, the approach presented herein consists of three stages: first, we use fuzzy TOPSIS to
rank the available suppliers on the basis of green criteria and traditional criteria separately, which results in two
preference weights (closeness coefficients) for each supplier: a green preference weight and a traditional
preference weight. We select fuzzy TOPSIS as an MCDM tool for ranking the suppliers because it performs
better in term of adequacy to changes of alternatives, adequacy to changes of criteria, agility in the decision
process, and number of criteria and suppliers, as shown in the comparison Lima Junior et al. [20] perform.
Fuzzy TOPSIS has been implemented using triangular fuzzy numbers (TFN). These numbers have different
advantages such as facilitating the ranking for the decision makers by using linguistic variables. Moreover,
TFNs capture the uncertainty of the knowledge of the decision makers as they allow for partial memberships,
whereas crisp sets allows for either full membership or non-membership. Second, we use AHP to determine an

9
importance weight for each of the two sets of criteria, so that if the company is focusing on its green
performance, the decision makers will be able to increase the relative importance weight of the set of green
criteria, independent of the available suppliers and their performance in terms of green criteria and traditional
criteria. While for the two categories of criteria the aim is to compare them according to the organization's
strategy, and because TOPSIS does not perform pairwise comparison, we select AHP for this stage. Applying
AHP for only two criteria and by a top management decision maker reduces the number of matrices required in
the calculation to one. This also eliminates the need to perform consistency checks because the results will be
completely consistent, as shown in section 3.2.1. Moreover, using only AHP for both levels (category and
criteria) would add more complexity in terms of number of matrices and may result in inconsistent rankings
because of the large number of criteria. Finally, because ranking the suppliers is not enough to select those with
which the company would contract, and because of other considerations, such as the capacity and the costs, the
third stage of our procedure uses optimization. We develop two optimization models—namely, a bi-objective
mathematical model and a multi-objective mathematical model—to select the best suppliers and determine the
number of products to be ordered in each period of a planning horizon. We detail the three stages in the next
sections.

3.1 Fuzzy TOPSIS

TOPSIS was introduced by Hwang and Yoon (1981). TOPSIS is an MCDM technique that ranks alternatives on
the basis of the shortest distance from the positive ideal solution and the longest distance from the negative ideal
solution [59]. Because decision makers prefer using a linguistic scale to numbers, and to account for vagueness
and uncertainty in the judgment procedure, Chen [60] proposed fuzzy TOPSIS. In the supplier selection
problem, fuzzy TOPSIS is preferred to fuzzy AHP, as shown in the comparison Lima Junior et al. [20] perform.
They proved that fuzzy TOPSIS generates consistent results when the number of alternatives or criteria changes,
while reversal ranking occurs in fuzzy AHP. They also showed that fuzzy TOPSIS performs better than fuzzy
AHP in most cases and requires less data. To implement fuzzy TOPSIS for the ranking of the suppliers based on
traditional criteria and green criteria, we use TFN.

3.1.1 TFN

Fuzzy theory was introduced by Zadeh [61] as an extension of the classic notion of sets. Fuzzy numbers are
powerful in describing a subjective measurement as a range, rather than as an exact value [62]. In the supplier
selection problem with a fuzzy environment, the most widely used format of fuzzy numbers is the TFN. A TFN
can be defined as a triplet (l, m, u), with a membership function defined as follows:

( ) { (1)

A TFN is based on a three-value judgment, where l is the minimum possible value, m is the most possible value,
and u is the maximum possible value.

A linguistic variable is a variable whose value is a word [63]. In this work, we use linguistic variables to account
for the uncertainty in the decision-making process because they are simple enough to be represented as fuzzy
numbers. Several types of linguistic scales for TFN were proposed and suggested with different points [62]. For
example, many researchers have proposed different five-point linguistic scales ([64], [65], [66], [67], [68], [69]).
Other researchers have proposed six-point linguistic scales ([70], [71]), while still others have proposed seven-
point linguistic scales ([72], [73], [74], [75]).

In this work, we use the five-point linguistic scale proposed by Lau et al. [68]. Tables 3 and 4 contain the
definition of our rating scale for criteria and alternatives (suppliers). Using these linguistic variables, every

10
decision maker will assign a weight to every criterion in the green set of criteria and to every criterion in the set
of traditional criteria. Moreover, every decision maker will assign a weight to every supplier with respect to
each criterion in each of the two sets of criteria. The assignment of the weights to the criteria is based on the
available knowledge and expertise of the decision makers as well as the relative importance of each criterion
within its set to the company. The assignment of weights to the suppliers with respect to the criteria can be done
using available historical data, the capability studies on the suppliers, and the laboratory testing and analysis of
the product to be purchased.

Table 3: Criterion rating scale.

Linguistic Variable TFN
Little importance (LI) (0, 0, 0.25)
Moderately important (MI) (0, 0.25, 0.50)
Important (I) (0.25, 0.50, 0.75)
Very important (VI) (0.50, 0.75, 1.00)
Absolutely important (AI) (0.75, 1.0, 1.0)
Table 4: Alternative rating scale.
Linguistic Variable TFN
Very low (VL) (0, 0, 0.25)
Low (L) (0, 0.25, 0.50)
Good (G) (0.25, 0.50, 0.75)
High (H) (0.50, 0.75, 1.00)
Very high (VH) (0.75, 1.0, 1.0)

3.1.2 Fuzzy TOPSIS procedure

Step 1:

Using linguistic variables, the decision maker , gives a linguistic weight transformed into a
TFN, , to each criterion , , and a linguistic weight transformed into TFN, , to each alternative,
, with respect to each criterion . In the current study, the alternatives are the available suppliers,
and the criteria are related to the suppliers and are either traditional or green. The weights are then aggregated
according to the following equations:

̅ ( ̅ ̅ ̅) ( ) (2)

̅ ( ̅ ̅ ̅ ) ( ) (3)

where ( ), a fuzzy number, is the criterion i weight given by decision maker Dk, and
( ), a fuzzy number, is the alternative j weight with respect to criterion i given by decision maker Dk.

Step 2:

In this step, we eliminate the different units of measurement using a normalization approach for the weights ̅
as follows:

̅ ̅ ̅
( ) (4)

for every benefit criterion , and

(̅ ̅ ̅
) (5)

11
for every cost criterion, , where is the normalized value of ̅ and and [17]. We
then combine the matrix [ ] with the vector [ ̅ ] to form the decision matrix.

Step 3:

In this step, each alternative weight must be multiplied by each criterion weight to obtain the weighted
normalized fuzzy decision matrix as in the following equation:

[ ] (6)

In other words, we obtain V by multiplying the last row in the decision matrix obtained in step 2, ̅ , by each
row of that matrix.

Step 4:

This step identifies the fuzzy positive ideal solution (FPIS) and the fuzzy negative ideal solution (FNIS). As the
normalized values of are from 0 to 1 [20,76], the FPIS is defined as (1,1,1), while the FNIS is defined as
(0,0,0). We use these values to calculate the distance from the positive ideal ( ) and the negative ideal ( )
solutions for each alternative as follows:

∑ ( ) (7)

∑ ( ) (8)

where we calculate and using

( ) √ [( ) ( ) ( ) ] (9)

where ( ) is the distance between ( ) and ( ).

Step 5:

This last step ranks the alternatives according to the closeness coefficient, , which we calculate
as follows:

(10)

3.2 AHP

In our model, we use the AHP to assign weights to the two sets of criteria—namely, the set of traditional criteria
and the set of green criteria. AHP is an MCDM tool that Saaty [77] developed in the 1970s. AHP decomposes
the problem into a hierarchical order, beginning with a goal, then main criteria, then sub-criteria, and, finally,
alternatives [77]. We explain the basic idea of AHP next.

3.2.1 AHP procedure for calculating weights of categories

12
 AHP

Goal

Criteria 1 Criteria 2

Alt 1 Alt 2 Alt 3 Alt 1 Alt 2 Alt 3

Fig. 1. AHP.
AHP consists mainly of three levels, as Fig. 1 shows: the goal, the criteria, and the alternatives. To achieve the
goal, the alternative weights are multiplied by each criterion weight and then added together to obtain the total
weight of the alternative. In this work, the goal of AHP is to calculate the importance weights of the two sets of
criteria, traditional criteria, and green criteria, by constructing a pairwise comparison between the two categories
of criteria. In each level of the hierarchy of AHP, we apply the following three steps:

Step 1:

Construct a pairwise comparison matrix for the criteria. Assign a relative weight, , for each of the two
criteria, i and j, by using the following scale: equally preferred (1), moderately preferred (3), strongly preferred
(5), very strongly preferred (7), and extremely preferred (9).

Step 2:

Normalize the values obtained from the pairwise comparison using the following equation:

̅ ∑(
(11)
)

where is the weight in the row and column of the criterion with respect to the criterion.

Step 3:

Calculate the weight, , for each criterion:

(12)

where is number of columns in the matrix.

We use the weights obtained from fuzzy TOPSIS using Eq. (10) and the weights obtained from AHP using Eq.
(12) as inputs to the two mathematical models described in the next sections.

Lemma 1:

The AHP procedure for ranking the green set of criteria and the traditional set of criteria is completely
consistent.

Proof 1:

The eigenvalue of matrix A can be calculated when the determinant of matrix equals 0, where I is the
identity matrix, A is the pairwise comparison matrix, and is the eigenvalue.

13
 |[ ] [ ]|

( )
| |
( )

( )( ) ( )

For the pairwise comparison herein, the two alternatives are the two sets of criteria. Therefore, n = 2. We
calculate the consistency index (CI) as follows:

which means that the consistency ratio, which is equal to CI/RI, is equal to 0,where RI is the random
consistency index; therefore, the pairwise comparison matrix is completely consistent and completes the proof.■

3.3 Bi-objective integer programming models

We use the bi-objective integer programming model to select the best suppliers and determine the quantities to
be ordered from each selected supplier in each period of a finite planning horizon. In addition, the mathematical
model will provide information about the optimal inventory levels at the end of each period.

The bi-objective optimization model consists of two objective functions. The first objective function aims to
minimize the TCP in terms of fixed cost, variable cost, inventory holding cost, and shortage cost over the
periods of the planning horizon. The second objective function tries to maximize total green criteria weights and
total traditional criteria weights over the same period. The model notations, objective functions, and constraint
formulations appear in the next sections.

3.3.1 Notations

Parameters:

 : Number of periods in the planning horizon.

 : Total number of suppliers.
 : Importance weight of the set of green criteria obtained from AHP.
 : Importance weight of the set of traditional criteria obtained from AHP.
 : Closeness coefficient obtained from fuzzy TOPSIS (Eq. (10)) for supplier ,
with respect to all the green criteria under consideration.
 : Closeness coefficient obtained from fuzzy TOPSIS (Eq. (10)) for supplier ,
with respect to all the traditional criteria under consideration.
 : Variable cost per unit product for supplier in period t, .
 : Fixed cost per order for supplier in period t, .
 : Inventory holding cost per unit product in period t, .
 : Shortage cost per unit product in period t, .
 : Supply capacity of supplier in period t, .

14
  : Minimum quantity to be ordered from supplier , if selected, in period t,
.
 : Deterministic demand in period t, .

Decision variables:

 : Quantity to be ordered from supplier j in period t, .

 : A binary variable for supplier j in period t that equals 1 if the supplier is selected and 0
otherwise, .
 : Available inventory level at the end of period t, .

3.3.2 Bi-objective integer non-linear programming model

In this section, we introduce the bi-objective integer non-linear programming (bi-objective INLP) model, using
the traditional way of modeling inventory levels in the backlog environment, which leads to a non-linear
formulation.

∑ ∑ ∑ ∑ (13)

∑ ∑ ( ( ) ( )) (14)

subject to:

(15)

∑ (16)

∑ ∑ ∑ (17)

)(18)

{ } (20)

Eq. (13) maximizes the total preference of all the selected suppliers over the planning horizon, defined as the
total value of purchasing (TVP). The first part of Eq. (13) refers to the green criteria, while the second part
refers to the general criteria, such as cost, delivery time, and quality. In this objective function, the weights of
the suppliers, which are based on the multiplication of the category weights obtained from AHP ( and
) by the respective supplier's preference weights obtained from fuzzy TOPSIS ( and ), are
multiplied by the quantities to be ordered from the supplier, to reflect the impact of the amount ordered on the
company’s performance. For example, when the solution selects two suppliers, one with high total combined
preference weight and one with low preference weight, the greater the quantity ordered from the preferred
supplier (with high preference weight), the better is the company’s performance. In contrast, ordering a small
amount from the supplier with low total combined preference weight results in a small effect on the overall
value of purchasing. The value of purchasing represents the total combined weights from the assessment
multiplied by the quantities ordered. Eq. (14) minimizes the TCP, which includes the variable cost, the fixed
cost, the inventory holding cost, and the penalty shortage cost. The variable cost is the cost per unit for a certain
product, while the fixed ordering cost is the cost of placing the order, inspection, and documentation. The
inventory holding cost is the cost associated with storing the product, which includes the capital cost, insurance
cost, obsolescence cost, storage keeper salary, and storage maintenance cost. Shortage cost is the penalty cost
for not satisfying the demand and delaying the order to a later period.

15
Constraints (15) ensure that each supplier’s capacity is not exceeded in each period and that the ordered quantity
is at least equal to the minimum specified value if the supplier is selected. Constraints (16) ensure that the
demand in each period is satisfied either from the quantity ordered or from the available inventory or is
backlogged to be satisfied in a later period. Constraint (17) ensures that the total demand for the entire planning
horizon is satisfied, though some demands may be satisfied in a later period than the period in which they occur.
Constraints (18) are for the integrality and non-negativity of the ordered quantities and the integrality of the
inventory levels, and constraints (19) ensure that the decision variables are binary.

We extend and reformulate this INLP, which was developed by Hamdan and Cheaitou [78], into a linear model
by introducing new decision variables and constraints, as shown in the next section.

3.3.3 Bi-objective ILP model

The decision variable, , is replaced with the following decision variables:

 : Available inventory level at the end of period t.

 : Shortage level at the end of period t.
 : A binary variable that indicates whether the inventory at the end of period t is positive
( ) or not ( ).
 : A binary variable that indicates whether the inventory at the end of period t is negative
( ) or not ( ).

In the new formulation, we keep the TVP objective function (Eq. (13)). In addition, we replace the total cost
objective function (Eq. (14)) by the following equation to eliminate the non-linear terms, that is, ( )
and ( ).

∑ ∑ ( ) (20)

In addition to constraints (15), (17), and (19), we add the following constraints:

{ } (21)

(22)

(23)

(24)

Moreover, in constraints (18), we replace by and .

Furthermore, we replace constraints (16) with the following constraints:

∑ (25)

Constraint (21) ensures that are binary variables. Constraint (22) ensures that if is negative, the
corresponding binary variable is 1, where and are a very large positive number and a very small positive
number, respectively. Constraint (23) ensures that if is positive, the corresponding binary variable is 1.
Constraint (24) ensures that at most one inventory type is chosen in each period—in other words, either positive
inventory, negative inventory, or neither. These modifications transform the non-linear model into a linear
model and make it possible to use the branch-and-cut algorithm implemented in the Optimization Toolbox of
MATLAB to solve the mathematical model.

3.4 Multi-objective ILP model

16
In the bi-objective optimization model described previously, both the total green preference weight and
traditional preference weight of the suppliers are taken into account in a single objective function (TVP). The
two preference weights of each supplier (closeness coefficients obtained from fuzzy TOPSIS) are multiplied by
the green and traditional criteria set weights (obtained from AHP), respectively, which results, as explained
previously, in a selection that depends on the decision maker’s preference for the two categories. An alternative
way that the decision maker can differentiate between the green and traditional criteria is by splitting the
objective function TVP into two separate objective functions, in which one represents the total green preference
weights of the suppliers and the other the total traditional preference weights. This method provides the decision
maker with more flexibility in finding the Pareto optimal solutions, as we explain in section 3.5, by varying the
weights given to the objective functions in the solution procedure. To explore the effect of considering the total
green value of purchasing (TGVP) and the total traditional value of purchasing (TTVP) in two objective
functions on the optimal solution, we introduce a multi-objective ILP model based on the bi-objective ILP
model defined in the previous section. The multi-objective model has the same parameters, decision variables,
and constraints as the bi-objective ILP model.

In contrast, the multi-objective configuration has three objective functions. The first objective function is the
one defined in Eq. (20) that minimizes the TCP. Moreover, the other two objective functions are obtained by
splitting Eq. (13), which maximizes the TVP in the bi-objective configuration, into the following two objective
functions:

∑ ∑ (26)

∑ ∑ (27)

Eq. (26) aims to maximize the TGVP, and Eq. (27) maximizes the TTVP.

The multi-objective optimization model is then defined by Eqs. (20), (26), and (27), subject to constraints (15),
(17), (19), (21)-(25), , and . Fig. 2 depicts the approach.

17
 Identification of the available suppliers by the procurement
department

Identification of the relevant criteria (green and traditional) by the

relevant decision makers and based on the product

Fuzzy TOPSIS Output

Calculation of the suppliers’ Calculation of the suppliers’ Suppliers’ green preference

preference weights based on preference weights based on weight and traditional
green criteria traditional criteria preference weight

×
AHP

Calculation of the importance weight of the sets of green and Green category weight and
traditional criteria traditional category weight

Optimization models

Bi-objective model Multi-objective model

Maximize TGVP (green value)

Maximize TVP (green and traditional values)
Maximize TTVP (traditional value)
Minimize TCP
Minimize TCP

Fig. 2. Summary of the MCDM and optimization approach.

3.5 Solution approach

In the presence of multiple and conflicting objectives, multiple solutions will exist because no one solution can
be the optimal solution for all the conflicting objectives. These solutions involve a set of non-dominated and
efficient optimal solutions called a Pareto-optimal set [79]. To obtain all the Pareto optimal solutions of the bi-
objective or multi-objective models described previously, scalarization techniques such as the weighted sum
method [80], the ε-constraint method [81], the comprehensive criterion method [82], or the WCCM [83,84] can
be used. In these techniques, some parameters can be changed to obtain the maximum number of Pareto
solutions. We adopt the WCCM to solve the bi-objective and multi-objective models because of the simplicity
of its implementation and its efficiency, in terms of the number of Pareto solutions.

3.5.1 WCCM

First, we divide the bi-objective ILP model defined in section 3.3.3 into two single objective sub-problems. The
first sub-problem is defined using Eq. (13) as the objective function, subject to the constraints defined in section
3.3.3 (Eqs. (15), (17), (19), (21)-(25) and the modified Eq. (18)). The second sub-problem has Eq. (20) as its
objective function, subject to the same constraints. We solve these two sub-problems separately to obtain their
optimal solutions, and respectively. Second, we adopt the WCCM to merge both objective
functions into a single normalized objective function [82]. The normalization approach calculates the
normalized objective functions as follows:

18
 (28)

(29)

Eqs. (28) and (29) represent the relative variations between the single objective functions TCP and TVP and
their corresponding optimal values. We then multiply the normalized objective functions by relative weights and
combine them in a single objective function as follows:

(30)

where and are the relative weights that the decision maker can set with . The total normalized
objective function (30) is then minimized, subject to the same constraints of the original problems (Eqs. (15),
(17), (19), (21)-(25) and the modified Eq. (18)). For given values of and , only one Pareto optimal solution
is obtained, while changing the values of and may result in different Pareto optimal solutions. Each
solution is non-dominated by any other solution [84].

The same approach can be applied to the multi-objective problem. Eq. (28) can still be used for the TCP
objective function, while Eq. (29) should be replaced by the following two equations for the TGVP and TTVP
objective functions as follows:

(31)

(32)

The combined normalized objective function is then given by

(33)

In Eq. (33), is equivalent to in Eq. (30), while , , and .

3.5.2 GUI

Using fuzzy TOPSIS, AHP, and then bi-objective optimization has been simplified as a procedure by
developing user-friendly software, as mentioned previously and shown in Figs. 3 and 4. The software is based
on a GUI of MATLAB, which asks decision makers to fill in their evaluation of the criteria and the alternatives
(suppliers). It also reads the input data (e.g., cost elements, supplier capacity, demand) from an MS Excel file. It
calculates the suppliers' closeness coefficients and the two sets of criteria importance weights and then feeds the
bi-objective optimization model that is solved using the branch-and-cut algorithm of the Optimization Toolbox
of MATLAB. The ranking and optimization results of the software are displayed in the GUI and stored in the
MS Excel file.

19
Fig. 3. Fuzzy TOPSIS and AHP tab in the software GUI.

Fig. 4. Optimization tab in the software GUI.

4. Numerical analysis
In this section, we first solve the bi-objective ILP model described previously using the WCCM method for
different numerical instances to analyze the effect of the model parameters and especially the effect of
separating the supplier selection criteria into two sets. Then, we compare this model with the multi-objective
configuration. The data used in this section are realistic experimental data that we randomly generated to test the
models. However, the models may be used with any other input data of any real case. Moreover, the models
have been implemented on a real-world case study, as shown in section 5.

4.1 Bi-objective model: basic numerical example

This section provides a numerical comparative study, based on the optimal solution, between the approach used
in this article and the "classic" approach adopted in the literature, in which all the supplier selection criteria
(green and traditional) are dealt with as a single set. The numerical study in this section is based on two
numerical instances: case 1 and case 2.
20
4.1.1 Input data for the mathematical model

For the mathematical model, we used the following data for the two case studies:

1. Six planning periods are considered (T = 6).

2. Five suppliers are considered (m = 5): S1, S2, S3, S4, and S5.
3. For case 1, three decision makers (DM1, DM2, and DM3) rank the suppliers on the basis of the green
and traditional criteria, while for case 2, only one decision maker (DM1) ranks the suppliers on the
basis of the different criteria.
4. The importance weights for the WCCM method are both equal to 0.5 ( 0.5).
5. The demand for all periods is equal to =1500 units per period for .
6. The minimum quantity to be ordered if supplier j is selected in period t is 20 units for

7. The fixed cost for every supplier is the same in all periods and is as follows: 3500 for supplier
1, 3380 for supplier 2, 3320 for supplier 3, 3275 for supplier 4, and
3250 for supplier 5, for .
8. The variable cost is the same in each period and is as follows: 55 per unit for supplier 1,
52 per unit for supplier 2, 50 per unit for supplier 3, 48 per unit for supplier 4, and
46 per unit for supplier 5, for .
9. The inventory holding cost is 10 per unit per period for all the periods in the planning horizon,
while the penalty shortage cost is 100 per unit per period.
10. Each supplier has a capacity of 10,000 units per period

4.1.2 Alternatives ranking

In both case 1 and case 2, we evaluate five suppliers on the basis of both traditional and green criteria using
fuzzy TOPSIS, in which we calculate a weight (closeness coefficient) for each alternative with respect to the
criteria in the set of traditional and green criteria. We also give the two sets of criteria importance weights using
AHP. Table 5 and Table 6 show the ranking inputs, for case 1 and case 2, respectively, where TRC represents a
traditional criterion (non-green), the asterisk (*) in the criterion name (TRC*) indicates that the criterion is a
negative criterion (i.e., cost and lead time), and GC represents green criterion. We assumed that there are three
decision makers for fuzzy TOPSIS and for case 1 and one for case 2. The input data shown in Tables 5 and 6 are
based on our assumptions.

Table 5: Ranking for case 1.

Criteria TRC*1 TRC2 TRC3 GC1 GC2 GC3
Criterion
DM 1 VI MI I VI I MI
rating
S1 VH VL G VH VH VH
S2 H L G H H G
Alternative
S3 G G VH G G VH
rating
S4 L H G L G G
S5 VL VH VH VL VL L
Criterion
DM2 I I I MI AI I
rating
S1 VH L H H H VH
S2 H G VH G VH G
Alternative
S3 G G L G VL G
rating
S4 L H VH VL L G
S5 VL H VH L VL VL
Criterion
DM3 AI LI I I I VI
rating
S1 VH VL L G H H
Alternative
S2 H L L VH G L
rating
S3 G G G G G G
21
 S4 L G VH VL G L
S5 VL H H VL L L

Table 6: Ranking for case 2.

Criteria TRC*1 TRC2 TRC3 TRC4 GC1 GC2 GC3

Criterion
DM1 VI MI I I VI I MI
rating
S1 VH L G G VH VH VH
S2 H G G L H H G
Alternative
S3 G VH G L G G VH
rating
S4 L H G VH L G G
S5 VL VH VH VH VL VL L

Tables 7 and 8 display the closeness coefficients of each supplier under two approaches for case 1 and case 2,
respectively: the approach based on fuzzy TOPSIS and AHP adopted in this article and the classic approach in
the literature, in which all the criteria are considered in a single set. In these tables, we assume different weights
for the two sets of criteria as the output from AHP, where G denotes green, T denotes traditional, and the
numbers next to G and T are the importance weights presumed to be obtained from AHP for the set of green
criteria and the set of traditional criteria, respectively. For example, 0.75G/0.25T means that and
. In the fuzzy TOPSIS and AHP approach, we apply fuzzy TOPSIS first for each supplier using the
green criteria and the traditional criteria separately. We then give different weights to each set of criteria (AHP).
The closeness coefficient of a supplier is then the combination of its closeness coefficients in each category and
the importance weights of these two categories (sets). Moreover, in the classic approach, we apply fuzzy
TOPSIS for each supplier, taking into account all the criteria (green and traditional) at the same time.

Table 7: Closeness coefficients for case 1.

Classic
Fuzzy TOPSIS +AHP
approach
Supplier
0.75G/0.25T 0.5G/0.5T 0.25G/0.75T
S1 0.1314 0.4824 0.3947 0.3069 0.2192 0.3191
S2 0.1608 0.4095 0.3473 0.2852 0.2230 0.2944
S3 0.2213 0.3405 0.3107 0.2809 0.2511 0.2664
S4 0.3290 0.2577 0.2755 0.2934 0.3112 0.3132
S5 0.4359 0.1600 0.2290 0.2980 0.3669 0.3039

As Table 7 shows, the weights in the classic approach of the literature correspond approximately to the case in
which the AHP weights for both categories are equal to 0.5 (0.5G/0.5T). This result confirms the importance of
the separation of the criteria into two categories.

Table 8: Closeness coefficients for case 2.

Classic
Fuzzy TOPSIS +AHP
approach
Supplier
0.75G/0.25T 0.5G/0.5T 0.25G/0.75T
S1 0.2035 0.4809 0.41155 0.3422 0.27285 0.3258
S2 0.1916 0.4127 0.357425 0.30215 0.246875 0.2896
S3 0.2126 0.352 0.31715 0.2823 0.24745 0.2736
S4 0.3779 0.2727 0.299 0.3253 0.3516 0.3346
S5 0.4563 0.1266 0.209025 0.29145 0.373875 0.3221

22
Table 8 shows that the closeness coefficients of the suppliers in the proposed approach (fuzzy TOPSIS + AHP),
with AHP weights equal to 0.5G/0.5T, are different from the closeness coefficients of the classic approach and
result in different supplier rankings. For example, S1 ranks first in the proposed approach (fuzzy TOPSIS +
AHP), while S4 ranks first in the classic approach.

4.1.3 Optimal solution

Tables 9–11 summarize the optimal solution of the bi-objective ILP model for the three different combinations
of criteria set weights as well as for the classic approach.

Table 9: Optimal quantities to order in case 1.

Fuzzy TOPSIS + AHP (0.75G/0.25T) Fuzzy TOPSIS + AHP (0.25G/0.75T)
Period 1 2 3 4 5 6 Period 1 2 3 4 5 6
S1 1500 1500 1500 1500 1500 1500 S1 0 0 0 0 0 0
S2 0 0 0 0 0 0 S2 0 0 0 0 0 0
S3 0 0 0 0 0 0 S3 0 0 0 0 0 0
S4 0 0 0 0 0 0 S4 0 0 0 0 0 0
S5 0 0 0 0 0 0 S5 1500 1500 1500 1500 1500 1500
Fuzzy TOPSIS + AHP (0.5G/0.5T) Classic approach
S1 0 0 0 0 0 0 S1 0 0 0 0 0 0
S2 0 0 0 0 0 0 S2 0 0 0 0 0 0
S3 0 0 0 0 0 0 S3 0 0 0 0 0 0
S4 0 0 0 0 0 0 S4 0 0 0 0 0 0
S5 1500 1500 1500 1500 1500 1500 S5 1500 1500 1500 1500 1500 1500

Table 10: Optimal quantities to order in case 2.

Fuzzy TOPSIS + AHP (0.75G/0.25T) Fuzzy TOPSIS + AHP (0.25G/0.75T)
Period 1 2 3 4 5 6 Period 1 2 3 4 5 6
S1 1500 1500 1500 1500 1500 1500 S1 0 0 0 0 0 0
S2 0 0 0 0 0 0 S2 0 0 0 0 0 0
S3 0 0 0 0 0 0 S3 0 0 0 0 0 0
S4 0 0 0 0 0 0 S4 0 0 0 0 0 0
S5 0 0 0 0 0 0 S5 1500 1500 1500 1500 1500 1500
Fuzzy TOPSIS + AHP (0.5G/0.5T) Classic approach
S1 0 0 0 0 0 0 S1 0 0 0 0 0 0
S2 0 0 0 0 0 0 S2 0 0 0 0 0 0
S3 0 0 0 0 0 0 S3 0 0 0 0 0 0
S4 1500 1500 1500 1500 1500 1500 S4 0 0 0 0 0 0
S5 0 0 0 0 0 0 S5 1500 1500 1500 1500 1500 1500

Table 11: Optimal values in case 1 and case 2.

and Optimal Optimal Optimal Optimal

TCP TVP TGVP TTVP
Case 1 0.75G/0.25T 516000 3551.00 4341.60 1182.60
0.5G/0.5T 433500 2681.55 1440.00 3923.10
Fuzzy
0.25G/0.75T 433500 3302.33 1440.00 3923.10
TOPSIS +
Case 2 0.75G/0.25T 516000 3703.95 4328.10 1831.50
AHP approach
0.5G/0.5T 451650 2927.70 2454.30 3401.10
0.25G/0.75T 433500 3364.88 1139.40 4106.70
Classic Case 1 NA 433500 2735.1 N/A N/A
approach Case 2 NA 433500 2898.9 N/A N/A

In case 1, the proposed approach with fuzzy TOPSIS and AHP, with AHP weights equal to 0.5G/0.5T, on the
one hand and the classic approach on the other hand provide the same optimal solution, as the closeness
coefficients of the two approaches are very close to each other. Conversely, in case 2, the optimal solution in the
proposed approach (fuzzy TOPSIS + AHP), with AHP weights equal to 0.5G/0.5T, differs from the optimal
23
solution in the classic approach because of the difference in the ranking of the suppliers between the two
approaches.

Moreover, in the 0.75G/0.25T combination, the selected suppliers in both case 1 and case 2 are the suppliers
with the highest green weight that results in the highest combined weight. In addition, in the 0.25G/0.75T
combination, in both case 1 and case 2, the selected suppliers are those with the highest combined weight and
the highest traditional weight.

4.1.4 Pareto front for case 1

Fig. 5 shows the different Pareto optimal solutions that have been obtained for case 1 with AHP weights equal
to 0.8G/0.2T. We obtained the solutions by solving the bi-objective ILP defined previously for case 1 using the
WCCM method and for different values of and . The combinations of values began with 0.05 and
0.95. Then, we increased and decreased by increments of 0.05 until 0.95 and 0.05. As
Fig. 5 shows, we obtained only four Pareto optimal solutions. Moreover, an improvement in the TVP occurs
with an increase in the TCP. The total increase of the TVP (from the least value to the greatest value), which is
equal to 91.56 %, causes an increase in the TCP of approximately 19.03%.

-1500

-2000

-2500
-TVP

-3000

-3500

-4000
420,000 440,000 460,000 480,000 500,000 520,000
TCP

Fig. 5. Pareto set and Pareto front for case 1.

4.1.5 Findings

This section illustrates how the proposed model can be applied step-by-step. We obtain the Pareto front by
changing the weights ( , ) in the WCCM method shown in Eq. (30). This section also compares the ranking
approach proposed herein, which applies fuzzy TOPSIS for traditional criteria and then again for green criteria
combined with AHP for the two sets of criteria taken apart, with the classic approach, in which all criteria are
considered in one category during the ranking. The results confirm that the suppliers' ranking and the
corresponding optimal solution using the classic approach that exists in the literature are a special case of the
results of the approach proposed in this article. We investigate this conclusion in more depth in the next
sections.

4.2 Comparison based on the ranking results

In this section, we provide a comparative study between the approach proposed in this article and the classic
approach adopted in the literature based on the suppliers’ ranking obtained from fuzzy TOPSIS. The motivation
for this comparison lies in the results of the comparison of the optimal solutions of the proposed approach
24
(fuzzy TOPSIS + AHP), with AHP weights equal to 0.5G/0.5T, with the classic approach observed in the
previous section.

Therefore, in this section, we conduct a numerical comparison between the classic approach of ranking
suppliers, which is based on placing both green and general (traditional) criteria in the same set, and the
approach proposed in this work, which suggests a separation between green and general (traditional) criteria.
The objective of the comparison is to explore the effect of separating the criteria into two sets on the ranking of
the suppliers and, consequently, on the optimal solutions.

We develop two numerical examples—namely, case 3 and case 4—with different numbers of traditional criteria
and green criteria. In case 3, we consider five green criteria and five traditional criteria, and in case 4, we
include two green criteria and eight traditional criteria. Only closeness coefficients of the suppliers for each
numerical example are shown in this section.

Table 12: Case 3 closeness coefficients.

Classic approach
Fuzzy TOPSIS + AHP
Supplier

0.75G/0.25T 0.5G/ 0.5T 0.25G/ 0.75T

S1 0.2903 0.4688 0.4242 0.3796 0.3349 0.3786
S2 0.2903 0.4095 0.3797 0.3499 0.3201 0.3493
S3 0.2492 0.3214 0.3034 0.2853 0.2673 0.2858
S4 0.3115 0.2065 0.2328 0.259 0.2853 0.2605
S5 0.3751 0.1317 0.1926 0.2534 0.3143 0.257

Table 13: Case 4 closeness coefficients.

Classic approach
Supplier Fuzzy TOPSIS + AHP

0.75G/0.25T 0.5G/0.5T 0.25G/0.75T

S1 0.4688 0.5552 0.5336 0.5120 0.4904 0.4859
S2 0.44 0.5729 0.5397 0.5065 0.4732 0.4665
S3 0.4031 0.4047 0.4043 0.4039 0.4035 0.4034
S4 0.3251 0.3881 0.3724 0.3566 0.3409 0.3377

Table 12 (case 3) shows that the results obtained by placing all the criteria in one category (the classic approach)
and the results obtained by separating green criteria from traditional criteria (the fuzzy TOPSIS and AHP
approach with ) are very close. This closeness verifies the results of case 1 discussed in the
previous section, in which the numbers of green and traditional criteria were equal. Moreover, the results of case
4 (Table 13), in which the number of traditional criteria is larger than the number of green criteria, show that the
classic approach provides closeness coefficients that are very close to the case in which only the traditional
criteria are considered. In this case, the effect of green criteria is reduced in the ranking process.

These results suggest that a relationship exists between the number of criteria in each category and the closeness
coefficients. More precisely,

(34)

where represents the number of green criteria, is the number of traditional criteria, is the total number
of criteria, is the weight of supplier j in the classic" approach, and and are the closeness
coefficients of supplier j with respect to green and traditional criteria taken separately.

25
Applying Eq. (34) to case 4 results in closeness coefficients that have very close values to those of the classic
approach, as Table 14 shows.

Table 14: Supplier closeness coefficients for Case 4.

Supplier
Classic approach for (0.2 G/0.8 T)
S1 0.4859 0.4861
S2 0.4665 0.4666
S3 0.4034 0.4034
S4 0.3377 0.3377

Figs. 6 and 7 show the closeness coefficients of the suppliers for case 3 and case 4 for five different cases: fuzzy
TOPSIS with traditional criteria only, fuzzy TOPSIS with green criteria only, fuzzy TOPSIS with AHP and
weights of 0.5G/0.5T and 0.2G/0.8T, and, finally, the classic approach. Figs. 6 and 7 confirm that the results
obtained using Eq. (34) are very close to the results of the classic approach and therefore confirm that the classic
approach is a special case of the approach proposed herein.

S1
0.5
0.4
0.3
Traditional criteria
0.2 only
S5 S2
0.1 Green criteria only
0
Classic approach

Fuzzy TOPSIS + AHP

with 0.5G/0.5T

S4 S3

Fig. 6. Radar chart for the closeness coefficients of Case 3.

26
 S1
0.6
0.5
0.4
0.3 Traditional criteria
only
0.2
Green criteria only
0.1
S4 0 S2
Classic apporach

Fuzzy TOPSIS + AHP

with 0.2G/0.8T

S3

Fig. 7. Radar chart for the closeness coefficients of case 4.

4.2.1 Findings

The main finding of the numerical experiments of this section is that the suppliers' ranking obtained from the
classic approach, in which all criteria are taken into account together, depends on the number of criteria that
exists in each category (traditional and green) in addition to the performance of each suppliers with respect to
each criteria. More precisely, we can obtain the ranking of the classic approach using Eq. (34), which depends
on the number of criteria in each category as well as the closeness coefficients of each category of criteria. This
equation indicates that the approach of the literature and the approach proposed herein give the same results in
one case. This case corresponds to the combination in which the importance weights of the two sets of criteria,
obtained from AHP, are equal to the ratio of the number of criteria in each set to the total number of criteria,
which means that the approach in the literature is a special case of the approach proposed in this article. For
example, if, on the one hand, the number of criteria in each category is equal ( ), the classic approach is
a special case that corresponds to the case in which . If, on the other hand, the number of
criteria in each category is different, the number of criteria in each category may affect the ranking of the
suppliers in the classic approach. In such a case, even if the company prefers to give more importance to green
criteria, a supplier with poor green performance may have an excellent ranking if the number of traditional
criteria is larger than the number of green criteria. This is not the case of the approach proposed herein.

4.3 Effect of the variable cost of the selected supplier on the solution

In this section, we choose a basic numerical example to examine the effect of increasing the variable cost of the
selected supplier on the solution and to find the break point, or the point after which the selected supplier is no
longer chosen in the optimal solution. The basic numerical example corresponds to the numerical example
presented in case 2 in section 4.1 with and (0.75G/0.25T). The optimal selected
supplier is supplier 1 (S1), which has the highest green value and a variable cost of 55 per unit for
.

The variable cost is increased with increments of one per unit of product to solve the optimization problem. We
plan to solve 21 instances, from a variable cost of 55 per unit to 75 per unit. Table 15 includes the
TCP and TVP for each instance (Eqs. (20) and (13), respectively).

Table 15 TCP and TVP for different values of the variable cost of supplier 1.
Instance TCP TVP Selected supplier

27
 Instance TCP TVP Selected supplier
1 55 516000 3703.95
2 56 525000 3703.95
Supplier 1
3 57 534000 3703.95
4 58 543000 3703.95
5 59 488280 3216.83
6 60 488280 3216.83
7 61 488280 3216.83
8 62 488280 3216.83
9 63 488280 3216.83
10 64 488280 3216.83
11 65 488280 3216.83
12 66 488280 3216.83
13 67 488280 3216.83 Supplier 2
14 68 488280 3216.83
15 69 488280 3216.83
16 70 488280 3216.83
17 71 488280 3216.83
18 72 488280 3216.83
19 73 488280 3216.83
20 74 488280 3216.83
21 75 488280 3216.83

Table 15 shows that the TCP increases with the increase of while the TVP does not change. The increase in
the TCP stops when supplier 1 is no longer selected, which can be expected.

Fig. 8 shows that supplier 1 (S1) is no longer the best supplier after the variable cost becomes equal to 59 per
unit, which we show by the sudden drop in both the TVP and the TCP. Instead of supplier 1, supplier 2 (S2)
becomes the best supplier with the lowest possible variation in terms of both TCP and TVP from their optimal
values (Eqs. (28) and (29), respectively) and, consequently, the minimum bi-objective function (Eq. (30))
compared with the other alternatives.

TCP TVP

550000 3800.00
540000 3700.00
530000 3600.00
520000 3500.00
510000 3400.00
TVP
TCP

500000 3300.00
490000 3200.00
480000 3100.00
470000 3000.00
460000 2900.00
55 57 59 61 63 65 67 69 71 73 75
Variable cost

Fig. 8. Changes in TCP and TVP while increasing the variable cost, , for supplier 1.
To better analyze this point of change, or the break point, at which the best supplier changes from S1 to S2, we
introduce the following equation, which calculates the total variation of the objective functions TCP and TVP
with respect to their optimal values for supplier j:

28
 ( ∑ ) ((∑ ) ( ))
(35)

Eq. (35) represents the total relative variation between the objective functions of the bi-objective model,
assuming that only supplier j is selected, and the optimal values of these objective functions. This equation
assumes that the capacity of supplier j in each period is higher than the demand of that period. To show the
usefulness of Eq. (35), we use the basic numerical example mentioned previously. For this basic example, the
optimal total purchasing cost , when solving the cost single objective function problem is 433,500,
while the optimal total purchasing value , when solving the preference weights single objective
function problem is 3,703.95.

By applying Eq. (35) for the five suppliers in the basic example, the total variations are 19.03%,
25.79%, 31.34%, 31.53%, and 49.21% for supplier 1 through supplier 5, respectively. The
results suggest that supplier 2 is the second-best supplier after supplier 1 in terms of the total variation between
the solution of the bi-objective model and the optimal solutions of each of the single objective models.
Therefore, when the total variation of supplier 1 becomes greater than the total variation of supplier 2, due to the
change of the variable cost, supplier 2 replaces supplier 1.

Therefore, to determine the exact value of the variable cost, , which corresponds to the break point,
should be determined using Eq. (35) for supplier 1, while keeping unknown, and should be made equal
to the total variation of supplier 2, which is also calculated using Eq. (35). The obtained equation is then solved
for . The obtained value of that corresponds to the break point is per unit. This result means
that if the variable cost per unit of supplier 1, , is greater than 58.25, supplier 1 is no longer the optimal
supplier to be selected, and supplier 2 replaces supplier 1. It is worth noting that this value of
calculated using Eq. (35) corresponds to the results shown in Fig. 8. Moreover, Table 16 includes the
variable cost corresponding to the break point for the different suppliers in the basic example, calculated using
Eq. (35). We calculate these values under the assumption that the corresponding supplier is selected as the
optimal one; consequently, they mean that the better suppliers have been excluded from the set of possible
suppliers.

Table 16: Variable cost break point for each supplier.

Supplier of the break point Excluded suppliers
S1 58.255 -
S2 54.741 S1
S3 50.094 S1, S2
S4 56.514 S1, S2, S3

4.3.1 Findings

The main finding of this section is that there is a variable cost threshold for the selected (optimal) supplier
beyond which the selected supplier will no longer be optimal and will be replaced by the second-best supplier.
This threshold can be calculated using Eq. (35) for every supplier.

4.4 Effect of the weights of the two sets of criteria

In this section, we conduct a numerical analysis on the weights of the green set and the traditional set of criteria,
and , using case 2 presented in section 4.1. We use 19 different combinations of weights,
beginning with 0.05 and 0.95 for the first combination. We increase the green category
weight, , by increments of 0.05 and decrease by increments of 0.05 until the weights become
equal to 0.95 and 0.05, respectively.

29
Table 17 summarizes the results. In the 0.4G/0.6T case, supplier 5, which has the least total cost, has an
aggregated closeness coefficient, = 0.3244, obtained from fuzzy TOPSIS
and AHP, while supplier 4 has a higher cost than supplier 5 but aggregated closeness coefficient 0.3358.
These values lead to total variations from the optimal solution of the single sub-problems equal to 3.39%
for supplier 5 and 4.19% for supplier 4; therefore, we select supplier 5 as the optimal choice. When
increases to 0.45, supplier 5 is no longer the second best in terms of the aggregated closeness
coefficient , as Table 18 shows, which increases its total variation from the optimal solution. Therefore, the
optimal solution moves from supplier 5 to supplier 4, as Table 17 illustrates.

Table 17: Optimal cost and value objective functions for case 2.
TCP TVP Selected (optimal) supplier
0.05G/0.95T 433500 3958.34 1139.40 4106.70
0.1G/0.9T 433500 3809.97 1139.40 4106.70
0.15G/0.85T 433500 3661.61 1139.40 4106.70
0.2G/0.8T 433500 3513.24 1139.40 4106.70
Supplier 5
0.25G/0.75T 433500 3364.88 1139.40 4106.70
0.3G/0.7T 433500 3216.51 1139.40 4106.70
0.35G/0.65T 433500 3068.15 1139.40 4106.70
0.4G/0.6T 433500 2919.78 1139.40 4106.70
0.45G/0.55T 451650 2975.04 2454.30 3401.10
0.5G/0.5T 451650 2927.70 2454.30 3401.10 Supplier 4
0.55G/0.45T 451650 2880.36 2454.30 3401.10
0.6G/0.4T 516000 3329.46 4328.10 1831.50
0.65G/0.35T 516000 3454.29 4328.10 1831.50
0.7G/0.3T 516000 3579.12 4328.10 1831.50
0.75G/0.25T 516000 3703.95 4328.10 1831.50
Supplier 1
0.8G/0.2T 516000 3828.78 4328.10 1831.50
0.85G/0.15T 516000 3953.61 4328.10 1831.50
0.9G/0.1T 516000 4078.44 4328.10 1831.50
0.95G/0.05T 516000 4203.27 4328.10 1831.50

Table 18: Closeness coefficients for case 2 at break point 1 ((0.4G/0.6T) and (0.45G/0.55T)).
0.4G/0.6T 0.45G/0.55T
Supplier Rank Rank
S1 0.4809 0.2035 0.3145 3 0.3283 2
S2 0.4127 0.1916 0.2800 4 0.2911 4
S3 0.3520 0.2126 0.2684 5 0.2753 5
S4 0.2727 0.3779 0.3358 1 0.3306 1
S5 0.1266 0.4563 0.3244 2 0.3079 3

Moreover, in the 0.55G/0.45T case, supplier 4 is the second highest in terms of the aggregated closeness
coefficient, , as Table 19 shows, but the second lowest in terms of cost, while supplier 1 is the best in terms
of the aggregated closeness coefficient but very expensive in terms of variable cost. This leads to a solution in
which supplier 4 is the optimal supplier because of its smaller variation from the optimal solution than supplier
1. However, in the case 0.6G/0.4T, the aggregated closeness coefficient of supplier 4 drops. This drop increases
the variation from the optimal single objective values, which results in supplier 1 having the smallest variation
and therefore being selected as the optimal supplier.

Table 19: Closeness coefficients for case 2 at the break point 2 ((0.55G/0.45T) and (0.6G/0.4T)).
0.55G/0.45T 0.6G/0.4T
Supplier Rank Rank
S1 0.4809 0.2035 0.3561 1 0.3699 1
S2 0.4127 0.1916 0.3132 3 0.3243 2
S3 0.3520 0.2126 0.2893 4 0.2962 4
S4 0.2727 0.3779 0.3200 2 0.3148 3
S5 0.1266 0.4563 0.2750 5 0.2585 5

30
 4500.00
Supplier 5 Supplier 4 Supplier 1

Total value of purchasing (TVP)

4000.00
3500.00
3000.00
2500.00
Break point 1 Break point 2
2000.00
1500.00
1000.00
500.00
0.00
0.15 G 0.85 T
0.05 G 0.95 T

0.25 G 0.75 T

0.35 G 0.65 T
0.40 G 0.60 T
0.45 G 0.55 T

0.55 G 0.45 T
0.60 G 0.40 T
0.65 G 0.35 T
0.70 G 0.30 T
0.75 G 0.25 T
0.80 G 0.20 T
0.85 G 0.15 T
0.90 G 0.10 T
0.95 G 0.05 T
0.1 G 0.9 T

0.2 G 0.8 T

0.3 G 0.7 T

0.50 G 0.5 T
Category weights

Fig. 9. TVP versus category weights in case 2.

Fig. 9 shows the decrease in the TVP in the first phase when supplier 5 is selected. The optimal solution then
switches to supplier 4 with a slight increase in the TVP, as the weight of supplier 4 is higher than the weight of
supplier 5. The TVP then begins to decrease until supplier 4 is replaced by supplier 1. Note that at the first break
point, supplier 4 is the supplier with the second-lowest cost and the best TVP, while supplier 1 has the best
green value. With the increase of , the chance of supplier 1 being chosen as the optimal supplier increases
until this happens at the second break point.

4.4.1 Findings

The importance weights assigned by top management to the sets of criteria (traditional and green) have a great
impact on the selected (optimal) suppliers. Changing these weights results in a change of the selected (optimal)
supplier. The change happens at some specific values (break points), as Fig. 9 shows. This numerical result
confirms the importance of considering the two categories of criteria in two distinct sets and allowing the
company’s top management to assign different importance weight to each set.

4.5 Comparison between the bi-objective and the multi-objective models

In this section, we compare the bi-objective ILP defined in section 3.3 with the multi-objective ILP defined in
section 3.4. We solve both models using the WCCM method defined in section 3.5. We test a numerical
example using 15 different weight scenarios for the WCCM method, as shown in Table 20.

Table 20: Weights scenarios for the WCCM method.

Run #
1 0.75 0.25 0.1 0.9 0.1 0.675 0.225
2 0.5 0.5 0.1 0.9 0.1 0.45 0.45
3 0.25 0.75 0.1 0.9 0.1 0.225 0.675
4 0.75 0.25 0.25 0.75 0.25 0.5625 0.1875
5 0.5 0.5 0.25 0.75 0.25 0.375 0.375
6 0.25 0.75 0.25 0.75 0.25 0.1875 0.5625
7 0.75 0.25 0.5 0.5 0.5 0.375 0.125
8 0.5 0.5 0.5 0.5 0.5 0.25 0.25
9 0.25 0.75 0.5 0.5 0.5 0.125 0.375
31
 10 0.75 0.25 0.75 0.25 0.75 0.1875 0.0625
11 0.5 0.5 0.75 0.25 0.75 0.125 0.125
12 0.25 0.75 0.75 0.25 0.75 0.0625 0.1875
13 0.75 0.25 0.9 0.1 0.9 0.075 0.025
14 0.5 0.5 0.9 0.1 0.9 0.05 0.05
15 0.25 0.75 0.9 0.1 0.9 0.025 0.075

We assume the following values and use them for the optimization models:

1. Six planning periods are considered (T = 6).

2. Three suppliers are considered (m = 3): S1, S2, and S3.
3. One decision maker (DM1) will rank the suppliers on the basis of the green and traditional criteria.
4. The demand for all periods is equal to = 5000 units per period for .
5. The minimum quantity to be ordered if supplier j is selected in period t is 20 units for

6. The fixed cost for every supplier is the same in all periods and is as follows: 1750 for supplier
1, 1650 for supplier 2, and 1550 for supplier 3, for .
7. The variable cost is the same in each period and is as follows: 52 per unit for supplier 1,
50 per unit for supplier 2, and 48 per unit for supplier 3, for .
8. The inventory holding cost is 10 per unit per period for all the periods in the planning horizon,
while the penalty shortage cost is 100 per unit per period.
9. Each supplier has a capacity of 10,000 units per period, where
10. The closeness coefficients of the suppliers obtained from fuzzy TOPSIS are as follows: 0.2903;
= 0.4688 for supplier 1, 0.2492; 0.3214 for supplier 2, and 0.3751;
0.1317 for supplier 3.

We first solve the bi-objective ILP and the multi-objective ILP using the WCCM method as explained in section
3.5. We then calculate the corresponding optimal TCP, TGVP, and TTVP using Eqs. (20), (26), and (27),
respectively. Table 21 shows the results for both the bi-objective and multi-objective models.

Table 21: Comparison between bi-objective and multi-objective results.

Bi-objective model Multi-objective model
Run
TCP TCP
1 1570500 14064 8709 1570500 14064 8709
2 1570500 14064 8709 1570500 14064 8709
3 1570500 14064 8709 1570500 14064 8709
4 1570500 14064 8709 1570500 14064 8709
5 1570500 14064 8709 1570500 14064 8709
6 1570500 14064 8709 1449300 3951 11253
7 1570500 14064 8709 1570500 14064 8709
8 1570500 14064 8709 1570500 14064 8709
9 1449300 3951 11253 1449300 3951 11253
10 1570500 14064 8709 1570500 14064 8709
11 1570500 14064 8709 1449300 3951 11253
12 1449300 3951 11253 1449300 3951 11253
13 1449300 3951 11253 1449300 3951 11253
14 1449300 3951 11253 1449300 3951 11253
15 1449300 3951 11253 1449300 3951 11253

Table 21 shows that in most of the cases, both models have the same optimal solutions and, consequently, the
same TCP, TGVP, and TTVP. The results confirm that the bi-objective model is preferred to the multi-objective
model. Indeed, in all the cases explored in which there is a difference in the optimal solution between the bi-
objective model and the multi-objective model, the bi-objective model yields a higher green value. Moreover,

32
the bi-objective model has a lower computation time because solving three optimization problems in the bi-
objective model is faster than solving four optimization problems in the multi-objective model.

Furthermore, the WCCM method gives more importance to the objective function with the highest importance
weight ( or ) assigned by the decision maker. As a result, an objective function with a very small importance
weight may have a very large variation from its optimal solution, with a small effect on the bi-objective or
multi-objective solution. For example, in the case shown in Table 21, the TGVP of the multi-objective optimal
solution results in a variation of 71.91% with respect to the TGVP of the green value single objective function
problem defined in Eq. (26). The variation is then multiplied by the weight of the objective function (0.1875),
which results in a variation of 13.48% in the multi-objective model. We solve other instances and confirm the
results. (These instances and their solutions are available on request.)

4.5.1 Findings

By comparing the bi-objective formulation and the multi-objective formulation, we show that in most of the
cases, there is no difference between the optimal solutions of the bi-objective and the multi-objective models.
Therefore, we recommend the bi-objective formulation, as it is simpler than the multi-objective one.

4.6 Computation time analysis for the bi-objective model

4.6.1 Computer specifications and computation time fitting

In this section, we analyze the total computation time required to solve the three optimization problems of the
bi-objective ILP model defined in section 3.3. All the optimization models herein have been solved using the
branch-and-cut algorithm implemented in the Optimization Toolbox of MATLAB R14 of MathWorks. The
models have been run on a computer equipped with an Intel(R) Core(TM) i5-4590, CPU @ 3.3 GHz 3.3 GHZ
and an 8.00 GB RAM and the windows 7 64-bit operating system.

We randomly generated the input data used in study for the mathematical models using the (RANDBETWEEN)
function in Microsoft Excel. We changed the problem size to examine the effect on the computation time by
increasing the number of suppliers and the number of periods from three suppliers and six periods to 40
suppliers and 250 periods, which leads to a maximum of 21,000 decision variables. We also solve the different
problems obtained and record their computation time. (The full data and results are available on request.)

Fig. 10 displays the total number of decision variables and constraints versus CPU running time. This plot can
be fitted as a second-degree polynomial function with a coefficient of determination , which means
that the obtained regression model accounts for 84% of the variability in the data. In Fig. 11, we fitted the
running time results with an exponential regression curve. After approximately 16,200 decision variables, the
computation time of the fitting curve increases rapidly. For example, the curve provides a computation time of
approximately 700,000 seconds for 20,000 decision variables, while the actual computation time for this number
of decision variables was less than 18.000 seconds. As a result, the exponential regression function, even with a
better coefficient of determination ( ) than the fitted polynomial function ( ), does not
reflect the behavior of the running time at a higher number of decision variables.

33
 25000
CPU running time (sec) y = 6E-06x2 - 0.0923x + 163.21
20000 R² = 0.8413

15000

10000

5000

0
0 20000 40000 60000 80000
Total number of decision variables and
constraints

Fig. 10. Total number of decision variables and constraints versus the CPU running time (polynomial fitting).

800000
700000
CPU running time (sec)

y = 2.3503e0.0006x
600000
R² = 0.8977
500000
400000
300000
200000
100000
0
0 5000 10000 15000 20000 25000
No. of decision variables

Fig. 11. Number of decision variables versus the CPU running time (exponential fitting).

4.6.2 Big-O notation

We use the big-O notation to classify an algorithm in terms of the required processing time as the input size
changes. A description of a function in terms of the big-O notation usually provides an upper bound on the
growth rate of the function [85]. The formal definition of the big-O notation is as follows:

( ) ( ( )) ( )

if and only if constants exist, such that ( ) ( ) for all , where is the problem
size, ( ) is the function that corresponds to the upper bound on the growth rate of the function ( ), and ( )
is the time function for the algorithm.

The big-O notation is based on two assumptions: the input argument, n, is restricted to be a non-negative integer
and the value of ( ) is non-negative for all values of n. As Fig. 10 shows, the best-fitting time function may
give negative values. To overcome this issue, we divided the data in Fig. 10 into two parts on the basis of the
number of decision variables. The first part represents the running time for the problems with fewer than 5,000

34
decision variables, and the second part represents the running time for the problems with 5,000 to 21,000
decision variables. Figs. 12 and 13 illustrate the results of this decision.

80 y = 4E-07x2 - 0.0017x + 3.4468

R² = 0.8797
70
CPU running time (sec)

60
50
40
30
20
10
0
0 5000 10000 15000 20000
No. of decision variables and constraints

Fig. 12. Total number of decision variables and constraints versus the CPU running time (part 1).

25000
y = 9E-06x2 - 0.2885x + 3250
R² = 0.772
CPU running time (sec)

20000

15000

10000

5000

0
0 20000 40000 60000 80000
No. of decision variables and constraints

Fig. 13. Total number of decision variables and constraints versus the CPU running time (part 2).
Figs. 12 and 13 show that the computation time function is polynomial and leads to the following lemma.

Lemma 2: As a function of the number of decision variables, n, the computation time, ( ), required to solve
the bi-objective ILP problem described in section 3.3 using the branch-and-cut algorithm has a polynomial
upper-bound function of ( ) ( ).

Proof 2:

For the best-fitting curve function in Fig. 12, consider ( ) .

Assuming , the goal is to find the value for c, such that represents the upper bound of the time
function, ( ) , which leads to

and, consequently, for gives

35
 Therefore, ( ) is ( ) and .

Moreover, for the best-fitting curve function in Fig. 13, consider ( )

. By assuming , we obtain

for all values of . Because the value of n in this equation is greater than or equal to 5,000, ( ) is
( ) for . This completes the proof. ■

4.6.3 Findings

The bi-objective model formulated in section 3.3.3 has a polynomial computation time function of ( )
( ). This result is useful because it confirms that the model can be used even for large problems with a large
number of possible suppliers and a large number of planning periods.

5. Case study
In this section, we use a case study from a facilities management company, which is based in the Gulf
Cooperation Council region and is a subsidiary of a multi-national company, to test the developed model and to
show its applicability using a realistic situation. We mention neither the company name nor the names of the
suppliers for confidentiality reasons; instead, we call the company XYZ and its potential three suppliers S1, S2,
and S3. The XYZ company wants to determine the optimum quantities of r141b refrigerant, which must be
purchased from its three certified green suppliers during a 12-month planning horizon. A decision maker of
XYZ used the MATLAB based software mentioned in Section 3.5.2 after having gone through the model and
confirmed its usefulness for this problem. The data related to the case study and the optimization results appear
in the following sections.

5.1 Data collection

XYZ ranks its suppliers on the basis of the following green aspects: compliance with government regulations
related to the environment, design for environment, environment protection, and supplier’s green image. XYZ
traditionally focuses on delivery time, payment terms, product cost, availability of stock, and quality. The
decision maker in the company was asked to assign an importance weight to each criterion and to evaluate each
supplier with respect to each criterion as shown in Table 22. The decision maker mentioned that the green aspect
is strongly preferred to these other traditional aspects. Table 23 also shows the categories’ importance weights
using AHP.

Table 22: Traditional and green ranking.

Delivery Payment Stock Compliance to Design for Environment Green
Criteria Cost Quality
time term availability government reg. environment protection image
Criterion
DM VI I I AI AI I VI VI I
rating
S1 H VH VH L VH G VH H H
Alternative
S2 G VH G G H L H L L
rating
S3 G VH G G H L H L L

36
 Table 23: Closeness coefficients and categories weights.
Category (set of criteria) importance
weights 0.1667 0.8333
Supplier
Supplier closeness S1 0.4806 0.5217
coefficients S2 0.5152 0.3397
S3 0.5152 0.3397

The product demand is 45 units per month for January–April, 90 units per month for May–September, 63 units
per month for October and December, and 72 units for November. Regarding the capacity, each supplier can
provide 30 units per month. The fixed cost in each month is 2,575 AED for supplier 1, 1,595 AED for supplier
2, and 1,565 AED for supplier 3. The variable cost in each month is 294 AED per unit for supplier 1 and 195
AED per unit for supplier 2 and supplier 3. The company does not allow for shortage in demand, because the
product is essential; thus, we set the shortage cost to a large value. The inventory holding cost for each unit in
each month is 30 AED.

5.2 Results

When only the TCP is minimized, the optimal TCP = 226717.6 AED. Conversely, when only the TVP is
maximized, the optimal TVP = 358.017. Finally, optimizing the bi-objective problem results in a total variation
from both optimal objectives equal to 9.18%. Table 24 summarizes the purchasing schedule in each step.

Table 24: Purchasing schedule for each optimization stage.

Optimal solution of the cost minimization sub-problem (TCP = 226717.6)

Jan. Feb. Mar. Apr. May Jun. Jul. Aug. Sep. Oct. Nov. Dec.
S1 0 0 0 0 0 0 30 30 30 18 0 0
S2 30 30 30 30 30 30 30 30 30 30 30 30
S3 30 30 30 30 30 30 30 30 30 30 30 30
Inventory 15 30 45 60 30 0 0 0 0 15 3 0
Shortage 0 0 0 0 0 0 0 0 0 0 0 0
Optimal solution of the TVP maximization sub-problem (TVP = 358.017)
S1 30 30 30 30 30 30 30 30 30 30 30 30
S2 15 15 15 0 30 30 30 30 30 30 30 30
S3 0 0 0 15 30 30 30 30 30 3 12 3
Inventory 0 0 0 0 0 0 0 0 0 0 0 0
Shortage 0 0 0 0 0 0 0 0 0 0 0 0
Optimal solution of the bi-objective problem (TCP = 227033, TVP = 325.629, variation = 9.18%)
S1 0 0 0 0 0 30 30 30 30 18 0 0
S2 30 0 30 30 30 30 30 30 30 30 30 30
S3 30 30 30 30 30 30 30 30 30 30 30 30
Inventory 15 0 15 30 0 0 0 0 0 15 3 0
Shortage 0 0 0 0 0 0 0 0 0 0 0 0

5.3 Discussion

Solving the TCP sub-problem leads to purchasing from supplier 2 and supplier 3 because they are cheaper than
supplier 1, with a difference in the variable cost of 99 AED per unit and almost 1,000 AED in the fixed cost.
Optimizing the TVP results in selecting supplier 1. This result is justified because supplier 1 has the highest
37
TVP per unit: 0.515 = 0.8333 × 0.5217 + 0.1667 × 0.4806. Considering both objectives leads to purchasing
from suppliers 2 and 3, mainly because the variation would increase significantly if supplier 1 were selected,
due to the larger difference in the costs than the difference in the weights among the three suppliers.

6. Conclusion
This article uses fuzzy TOPSIS to rank potential suppliers on the basis of two sets of criteria: traditional and
green. It then uses AHP to assign importance weights to each set of criteria. Consequently, the potential supplier
considered very good in terms of traditional criteria and very poor in terms of green criteria will not be ranked
among the best alternatives if top management decides to give more importance to the set of green criteria. This
model gives the decision maker more flexibility in deciding on the importance of the green aspect in the
selection of suppliers, which means that the decision maker can emphasize either set of criteria according to the
managerial perspective. We compared this proposed ranking approach with the classic approach adopted in the
literature in which all the criteria are considered in a single set. The ranking outcomes and the optimization
results show that the classic approach and the approach proposed herein give the same results in one case. This
case corresponds to the combination in which the importance weights of the two sets of criteria, for the proposed
approach, are equal to the ratio of the number of criteria in each set to the total number of criteria, which means
that the classic approach is a special case of the proposed approach. Furthermore, we conducted a sensitivity
analysis for the variable cost, in which we increased the variable cost of the selected supplier to obtain the break
point at which the optimal supplier changes. In addition, time analysis for the bi-objective ILP shows that the
developed model has a second-degree polynomial best-fitting function and that ( ) can be considered the
upper bound for the time function of the proposed model. Moreover, we compared the multi-objective
configuration with the bi-objective configuration under the same weights. The comparison results show that
both configurations lead to the same optimal solution in almost all the cases. This suggests that the bi-objective
model is superior to the multi-objective model because optimizing three problems in the bi-objective
configuration is faster than optimizing four problems in the multi-objective configuration. Finally, considering
stochastic demand and different potential suppliers in each period, as well as quantity discount schemes,
constitutes avenues for further research.

References
[1] Mafakheri F, Breton M, Ghoniem A. Supplier selection-order allocation: A two-stage multiple criteria
dynamic programming approach. Int J Prod Econ 2011;132:52–7. doi:10.1016/j.ijpe.2011.03.005.

[2] Guo C, Li X. A multi-echelon inventory system with supplier selection and order allocation under
stochastic demand. Int J Prod Econ 2014;151:37–47. doi:10.1016/j.ijpe.2014.01.017.

[3] Viswanadham N, Gaonkar RS, Member S. Partner Selection and Synchronized Planning in Dynamic
Manufacturing Networks 2003;19:117–30.

[4] Aissaoui N, Haouari M, Hassini E. Supplier selection and order lot sizing modeling: A review. Comput
Oper Res 2007;34:3516–40. doi:10.1016/j.cor.2006.01.016.

[5] Yeh W-C, Chuang M-C. Using multi-objective genetic algorithm for partner selection in green supply
chain problems. Expert Syst Appl 2011;38:4244–53. doi:10.1016/j.eswa.2010.09.091.

[6] Shaw K, Shankar R, Yadav SS, Thakur LS. Supplier selection using fuzzy AHP and fuzzy multi-
objective linear programming for developing low carbon supply chain. Expert Syst Appl 2012;39:8182–
92. doi:10.1016/j.eswa.2012.01.149.

[7] Igarashi M, de Boer L, Fet AM. What is required for greener supplier selection? A literature review and
conceptual model development. J Purch Supply Manag 2013;19:247–63.
doi:10.1016/j.pursup.2013.06.001.

38
[8] Büyüközkan G, Çifçi G. A novel hybrid MCDM approach based on fuzzy DEMATEL, fuzzy ANP and
fuzzy TOPSIS to evaluate green suppliers. Expert Syst Appl 2012;39:3000–11.
doi:10.1016/j.eswa.2011.08.162.

[9] Mendoza A, Ventura JA. A serial inventory system with supplier selection and order quantity allocation.
Eur J Oper Res 2010;207:1304–15. doi:10.1016/j.ejor.2010.06.034.

[10] Ware NR, Singh SP, Banwet DK. A mixed-integer non-linear program to model dynamic supplier
selection problem. Expert Syst Appl 2014;41:671–8. doi:10.1016/j.eswa.2013.07.092.

[11] Faez F, Ghodsypour SH, O’Brien C. Vendor selection and order allocation using an integrated fuzzy
case-based reasoning and mathematical programming model. Int J Prod Econ 2009;121:395–408.
doi:10.1016/j.ijpe.2006.11.022.

[12] Govindan K, Rajendran S, Sarkis J, Murugesan P. Multi criteria decision making approaches for green
supplier evaluation and selection: a literature review. J Clean Prod 2013.
doi:10.1016/j.jclepro.2013.06.046.

[13] Kannan D, Khodaverdi R, Olfat L, Jafarian A, Diabat A. Integrated fuzzy multi criteria decision making
method and multi-objective programming approach for supplier selection and order allocation in a green
supply chain. J Clean Prod 2013;47:355–67. doi:10.1016/j.jclepro.2013.02.010.

[14] Zhang J, Zhang M. Supplier selection and purchase problem with fixed cost and constrained order
quantities under stochastic demand. Int J Prod Econ 2011;129:1–7. doi:10.1016/j.ijpe.2010.08.003.

[15] Özgen D, Önüt S, Gülsün B, Tuzkaya UR, Tuzkaya G. A two-phase possibilistic linear programming
methodology for multi-objective supplier evaluation and order allocation problems. Inf Sci (Ny)
2008;178:485–500. doi:10.1016/j.ins.2007.08.002.

[16] Sanayei A, Farid Mousavi S, Abdi MR, Mohaghar A. An integrated group decision-making process for
supplier selection and order allocation using multi-attribute utility theory and linear programming. J
Franklin Inst 2008;345:731–47. doi:10.1016/j.jfranklin.2008.03.005.

[17] Kilic HS. An integrated approach for supplier selection in multi-item/multi-supplier environment. Appl
Math Model 2013;37:7752–63. doi:10.1016/j.apm.2013.03.010.

[18] Kannan D, Jabbour ABLDS, Jabbour CJC. Selecting green suppliers based on GSCM practices: Using
fuzzy TOPSIS applied to a Brazilian electronics company. Eur J Oper Res 2014;233:432–47.
doi:10.1016/j.ejor.2013.07.023.

[19] Amindoust A, Ahmed S, Saghafinia A, Bahreininejad A. Sustainable supplier selection: A ranking

model based on fuzzy inference system. Appl Soft Comput J 2012;12:1668–77.
doi:10.1016/j.asoc.2012.01.023.

[20] Lima Junior FR, Osiro L, Carpinetti LCR. A comparison between Fuzzy AHP and Fuzzy TOPSIS
methods to supplier selection. Appl Soft Comput 2014;21:194–209. doi:10.1016/j.asoc.2014.03.014.

[21] Razmi J, Maghool E. Multi-item supplier selection and lot-sizing planning under multiple price
discounts using augmented ε-constraint and Tchebycheff method. Int J Adv Manuf Technol 2009;49.
doi:10.1007/s00170-009-2392-1.

[22] Wu WY, Sukoco BM, Li CY, Chen SH. An integrated multi-objective decision-making process for
supplier selection with bundling problem. Expert Syst Appl 2009;36:2327–37.
doi:10.1016/j.eswa.2007.12.022.

[23] Dickson GW. An analysis of vendor selection systems and decisions. J Purch 1966;2:5–17.
doi:10.5465/AMBPP.1966.4980919.

[24] Chou S, Chang Y. A decision support system for supplier selection based on a strategy-aligned fuzzy
SMART approach. Expert Syst Appl 2008;34:2241–53. doi:10.1016/j.eswa.2007.03.001.

39
[25] Wang J-W, Cheng C-H, Huang K-C. Fuzzy hierarchical TOPSIS for supplier selection. Appl Soft
Comput 2009;9:377–86. doi:10.1016/j.asoc.2008.04.014.

[26] Kwong CK, Ip WH, Chan JWK. Combining scoring method and fuzzy expert systems approach to
supplier assessment: a case study. Integr Manuf Syst 2002;13:512–9. doi:10.1108/09576060210442671.

[27] Bruno G, Esposito E, Genovese A, Simpson M. Applying supplier selection methodologies in a multi-
stakeholder environment: A case study and a critical assessment. Expert Syst Appl 2016;43:271–85.
doi:10.1016/j.eswa.2015.07.016.

[28] Tavana M, Fallahpour A, Caprio D Di, Santos-Arteaga FJ. A Hybrid Intelligent Fuzzy Predictive Model
with Simulation for Supplier Evaluation and Selection. Expert Syst Appl 2016.
doi:10.1016/j.eswa.2016.05.027.

[29] Rahiminezhad M, Ahmad S, Hashemzahi P. Supplier selection in automobile industry  : A mixed

balanced scorecard – fuzzy AHP approach. Alexandria Eng J 2016;55:93–100.
doi:10.1016/j.aej.2016.01.005.

[30] Memon MS, Lee YH, Mari SI. Group multi-criteria supplier selection using combined grey systems
theory and uncertainty theory. Expert Syst Appl 2015;42:7951–9. doi:10.1016/j.eswa.2015.06.018.

[31] Shemshadi A, Shirazi H, Toreihi M, Tarokh MJ. A fuzzy VIKOR method for supplier selection based
on entropy measure for objective weighting. Expert Syst Appl 2011;38:12160–7.
doi:10.1016/j.eswa.2011.03.027.

[32] Du B, Guo S, Huang X, Li Y, Guo J. A Pareto supplier selection algorithm for minimum the life cycle
cost of complex product system. Expert Syst Appl 2015;42:4253–64. doi:10.1016/j.eswa.2015.01.056.

[33] Karpak B, Kumcu E, Kasuganti R. An application of visual interactive goal programming: a case in
vendor selection decisions. J Multicriteria Decis Anal 1999;8:93–105.

[34] Karpak B, Kasuganti RR, Kumcu E. Multi-objective decision-making in supplier selection: An

application of visual interactive goal programming. J Appl Bus Res 1999;15:57–72.

[35] Karpak B, Kumcu E, Kasuganti RR. Purchasing materials in the supply chain: Managing a multi-
objective task. Eur J Purch Supply Manag 2001;7:209–16. doi:10.1016/S0969-7012(01)00002-8.

[36] Kumar M, Vrat P, Shankar R. A fuzzy goal programming approach for vendor selection problem in a
supply chain. Comput Ind Eng 2004;46:69–85. doi:10.1016/j.cie.2003.09.010.

[37] Amorim P, Curcio E, Almada-Lobo B, Barbosa-Póvoa APFD, Grossmann IE. Supplier Selection in the
Processed Food Industry under Uncertainty. Eur J Oper Res 2016;252:801–14.
doi:10.1016/j.ejor.2016.02.005.

[38] Moghaddam KS. Fuzzy multi-objective model for supplier selection and order allocation in reverse
logistics systems under supply and demand uncertainty. Expert Syst Appl 2015;42:6237–54.
doi:10.1016/j.eswa.2015.02.010.

[39] Amin SH, Razmi J, Zhang G. Supplier selection and order allocation based on fuzzy SWOT analysis
and fuzzy linear programming. Expert Syst Appl 2011;38:334–42. doi:10.1016/j.eswa.2010.06.071.

[40] Xia W, Wu Z. Supplier selection with multiple criteria in volume discount environments 2007;di:494–
504. doi:10.1016/j.omega.2005.09.002.

[41] Demirtas EA, Üstün Ö. An integrated multiobjective decision making process for supplier selection and
order allocation. Omega 2008;36:76–90. doi:10.1016/j.omega.2005.11.003.

[42] Lee AHI, Kang HY, Chang CT. Fuzzy multiple goal programming applied to TFT-LCD supplier
selection by downstream manufacturers. Expert Syst Appl 2009;36:6318–25.
doi:10.1016/j.eswa.2008.08.044.

40
[43] Liao C-N, Kao H-P. An integrated fuzzy TOPSIS and MCGP approach to supplier selection in supply
chain management. Expert Syst Appl 2011;38:10803–11. doi:10.1016/j.eswa.2011.02.031.

[44] Rouyendegh BD, Saputro TE. Supplier Selection Using Integrated Fuzzy TOPSIS and MCGP: A Case
Study. Procedia - Soc Behav Sci 2014;116:3957–70. doi:10.1016/j.sbspro.2014.01.874.

[45] Perçin S. An application of the integrated AHP-PGP model in supplier selection. Meas Bus Excell
2006;10:34–49. doi:10.1108/13683040610719263.

[46] Ghorbani M, Bahrami M, Arabzad SM. Implementing Shannon Entropy, SWOT and Mathematical
Programming for Supplier Selection and Order Allocation. Int J Supply Chain Manag 2012;1:43–7.
doi:10.1016/j.sbspro.2012.04.064.

[47] Sodenkamp MA, Tavana M, Di Caprio D. Modeling synergies in multi-criteria supplier selection and
order allocation: An application to commodity trading. Eur J Oper Res 2016;0:1–16.
doi:10.1016/j.ejor.2016.04.015.

[48] Genovese A, Koh SCL, Bruno G, Bruno P. Green supplier selection: A literature review and a critical
perspective. Supply Chain Manag. Inf. Syst., 2010, p. 1–6. doi:978-962-367-696-0.

[49] Kannan D, Govindan K, Rajendran S. Fuzzy Axiomatic Design approach based green supplier selection:
a case study from Singapore. J Clean Prod 2014;96:-.
doi:http://dx.doi.org/10.1016/j.jclepro.2013.12.076.

[50] Hsu C-W, Kuo T-C, Chen S-H, Hu AH. Using DEMATEL to develop a carbon management model of
supplier selection in green supply chain management. J Clean Prod 2013;56:164–72.
doi:10.1016/j.jclepro.2011.09.012.

[51] Freeman J, Chen T. Green supplier selection using an AHP-Entropy-TOPSIS framework. Supply Chain
Manag An Int J 2015;20:327–40. doi:10.1108/SCM-04-2014-0142.

[52] Darabi S, Heydari J. An Interval- Valued Hesitant Fuzzy Ranking Method based on Group Decision
Analysis for Green Supplier Selection. IFAC-PapersOnLine 2016;49:12–7.
doi:10.1016/j.ifacol.2016.03.003.

[53] Galankashi MR, Chegeni A, Soleimanynanadegany A, Memari A, Anjomshoae A, Helmi SA, et al.
Prioritizing Green Supplier Selection Criteria Using Fuzzy Analytical Network Process. Procedia CIRP
2015;26:689–94. doi:10.1016/j.procir.2014.07.044.

[54] Hashemi SH, Karimi A, Tavana M. An integrated green supplier selection approach with analytic
network process and improved Grey relational analysis. Int J Prod Econ 2015;159:178–91.
doi:10.1016/j.ijpe.2014.09.027.

[55] Dobos I, Vörösmarty G. Green supplier selection and evaluation using DEA-type composite indicators.
Int J Prod Econ 2014;157:273–8. doi:10.1016/j.ijpe.2014.09.026.

[56] Zouggari A, Benyoucef L. Simulation based fuzzy TOPSIS approach for group multi-criteria supplier
selection problem. Eng Appl Artif Intell 2012;25:507–19. doi:10.1016/j.engappai.2011.10.012.

[57] Li Z, Wong WK, Kwong CK. An Integrated Model of Material Supplier Selection and Order Allocation
Using Fuzzy Extended AHP and Multiobjective Programming. Math Probl Eng 2013;2013:1–14.
doi:10.1155/2013/363718.

[58] Lee AHI, Kang H-Y, Lai C-M, Hong W-Y. An integrated model for lot sizing with supplier selection
and quantity discounts. Appl Math Model 2013;37:4733–46. doi:10.1016/j.apm.2012.09.056.

[59] Hwang C-L, Yoon K. Methods for Multiple Attribute Decision Making. Mult. Attrib. Decis. Mak. SE -
3, vol. 186, Springer Berlin Heidelberg; 1981, p. 58–191. doi:10.1007/978-3-642-48318-9_3.

[60] Chen C-T. Extensions of the TOPSIS for group decision-making under fuzzy environment. Fuzzy Sets
Syst 2000;114:1–9. doi:10.1016/S0165-0114(97)00377-1.
41
[61] Zadeh LA. 1965 J(Zadeh) Fuzzy Sets.pdf. Inf Control 1965;8:338–53. doi:10.1109/2.53.

[62] Chen T-Y, Ku T-C. Importance-assessing method with fuzzy number-valued fuzzy measures and
discussions on TFNs and TrFNs. Int J Fuzzy Syst 2008;10:92.

[63] Zadeh LA. The Concept of a Linguistic Variable and its Applications to Approximate Reasoning I. Inf
Sci (Ny) 1975;8:199–249.

[64] Chang P-L, Chen Y-C. A fuzzy multi-criteria decision making method for technology transfer strategy
selection in biotechnology. Fuzzy Sets Syst 1994;63:131–9. doi:10.1016/0165-0114(94)90344-1.

[65] Liou TS, Wang MJJ. Subjective Assessment of Mental Workload - a Fuzzy Linguistic Multicriteria
Approach. Fuzzy Sets Syst 1994;62:155–65. doi:Doi 10.1016/0165-0114(94)90055-8.

[66] Chien C-J, Tsai H-H. Using fuzzy numbers to evaluate perceived service quality. Fuzzy Sets Syst
2000;116:289–300. doi:10.1016/S0165-0114(98)00239-5.

[67] Lee HT, Chen SH, Kang HY. Multicriteria scheduling using fuzzy theory and tabu search. Int J Prod
Res 2002;40:1221–34. doi:10.1080/00207540110098832.

[68] Lau HCW, Wong CWY, Lau PKH, Pun KF, Chin KS, Jiang B. A fuzzy multi-criteria decision support
procedure for enhancing information delivery in extended enterprise networks. Eng Appl Artif Intell
2003;16:1–9. doi:10.1016/S0952-1976(03)00020-4.

[69] Yang T, Hung C-C. Multiple-attribute decision making methods for plant layout design problem. Robot
Comput Integr Manuf 2007;23:126–37. doi:10.1016/j.rcim.2005.12.002.

[70] Chan FTS, Chan MH, Tang NKH. Evaluation methodologies for technology selection. J Mater Process
Technol 2000;107:330–7. doi:10.1016/S0924-0136(00)00679-8.

[71] Ertay T, Büyüközkan G, Kahraman C, Ruan D. Quality function deployment implementation based on
analytic network process with linguistic data: An application in automotive industry. J Intell Fuzzy Syst
2005;16:221–32.

[72] Chen C-T. A fuzzy approach to select the location of the distribution center. Fuzzy Sets Syst
2001;118:65–73. doi:10.1016/S0165-0114(98)00459-X.

[73] Herrera F, Martínez L, Sánchez PJ. Managing non-homogeneous information in group decision making.
Eur J Oper Res 2005;166:115–32. doi:10.1016/j.ejor.2003.11.031.

[74] Chen C-T, Lin C-T, Huang S-F. A fuzzy approach for supplier evaluation and selection in supply chain
management. Int J Prod Econ 2006;102:289–301. doi:10.1016/j.ijpe.2005.03.009.

[75] Kuo M-S, Liang G-S, Huang W-C. Extensions of the multicriteria analysis with pairwise comparison
under a fuzzy environment. Int J Approx Reason 2006;43:268–85. doi:10.1016/j.ijar.2006.04.006.

[76] Saghafian S, Hejazi SR. Multi-criteria Group Decision Making Using A Modified Fuzzy TOPSIS
Procedure. Int Conf Comput Intell Model Control Autom Int Conf Intell Agents, Web Technol Internet
Commer 2005;2. doi:10.1109/CIMCA.2005.1631471.

[77] Saaty TL. Fundamentals of decision making and priority theory with the analytic hierarchy process. vol.
6. Rws Publications; 2000.

[78] Hamdan S, Cheaitou A. Green Supplier Selection and Order Allocation Using an Integrated Fuzzy
TOPSIS, AHP and IP Approach. Ind. Eng. Oper. Manag. (IEOM), 2015 Int. Conf., Dubai: IEEE; 2015,
p. 1–10. doi:10.1109/IEOM.2015.7093826.

[79] Fonseca CM, Fleming PJ. Multiobjective optimization and multiple constraint handling with
evolutionary algorithms - Part I: A unified formulation. IEEE Trans Syst Man, Cybern Part ASystems
Humans 1998;28:26–37. doi:10.1109/3468.650319.

42
[80] Miettinen K. Nonlinear multiobjective optimization. Int Ser Oper Res Manag Sci 1999;12:320.
doi:10.2307/254267.

[81] Mavrotas G. Effective implementation of the epsilon-constraint method in Multi-Objective

Mathematical Programming problems. Appl Math Comput 2009;213:455–65.
doi:10.1016/j.amc.2009.03.037.

[82] Dehghani M, Esmaeilian M, Tavakkoli-Moghaddam R. Employing Fuzzy ANP for Green Supplier
Selection and Order Allocations: A Case Study . Int J Econ Manag Soc Sci 2013;2:565–75.

[83] Kamali A, Fatemi Ghomi SMT, Jolai F. A multi-objective quantity discount and joint optimization
model for coordination of a single-buyer multi-vendor supply chain. Comput Math with Appl
2011;62:3251–69. doi:10.1016/j.camwa.2011.08.040.

[84] Marler RT, Arora JS. Survey of multi-objective optimization methods for engineering. Struct
Multidiscip Optim 2004;26:369–95. doi:10.1007/s00158-003-0368-6.

[85] Leiserson CCE, Rivest RRL, Stein C, Cormen TH. Introduction to Algorithms. 3rd ed. Massachusetts:
The MIT Press; 2009. doi:10.2307/2583667.

Highlights:

 Green and traditional criteria for supplier selection were split into two sets
 Two bi-objective and one multi-objective optimization models were proposed
 The models use a combination of three tools: AHP, Fuzzy TOPSIS and optimization
 The results show that the proposed approach is more flexible than the existing ones

43