You are on page 1of 10

Expert Systems with Applications 37 (2010) 3943–3952

Contents lists available at ScienceDirect

Expert Systems with Applications


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

Multi-criteria warehouse location selection using Choquet integral


Tufan Demirel a, Nihan Çetin Demirel a, Cengiz Kahraman b,*
a
Yıldız Technical University, Department of Industrial Engineering, Yıldız, Istanbul, Turkey
b
Istanbul Technical University, Department of Industrial Engineering, Maçka, Istanbul, Turkey

a r t i c l e i n f o a b s t r a c t

Keywords: The location of a warehouse is generally one of the most important and strategic decision in the optimi-
Choquet integral zation of logistic systems. Warehouse location is a long-term decision and is influenced by many quan-
Fuzzy titative and qualitative factors. Among the main criteria taken into account in this paper, some are costs,
Hierarchy labor characteristics, infrastructure, and markets. This paper also includes some sub-criteria because of
Multi-criteria
the hierarchical structure of the problem, like tax incentives and tax structures, availability of labor force,
Warehouse location
Fuzzy measure
quality and reliability of modes of transportation, and proximity to customers. The conventional
approaches to warehouse location selection problem tend to be less effective in dealing with the impre-
cise or vague nature of the linguistic assessment. Under many situations, the values of the qualitative cri-
teria are often imprecisely defined for the decision-makers. Choquet integral is a suitable multi-criteria
method to capture this imprecise or vague nature. This paper shows a successful application of multi-cri-
teria Choquet integral to a real warehouse location selection problem of a big Turkish logistic firm.
Crown Copyright Ó 2009 Published by Elsevier Ltd. All rights reserved.

1. Introduction system (PWSS) software has been built by Colson and Dorigo
(2004) to give the opportunity to industrial users of exploiting a
Optimum location of distribution warehouses is an important classical data base on public warehouses, where several items of
partial task of optimization of logistic systems. The efficient and information are given on each warehouse located in a given coun-
effective movement of goods from raw material sites to processing try. Their software public warehouses selection support has two
facilities, component fabrication plants, finished goods assembly purposes: to select public warehouses according to several criteria
plants, distribution centers, warehouses, retailers, and customers and to exploit a database when some data are missing. They use
is critical in today’s competitive environment. In today’s rapidly multiple criteria selections and rankings with a mixture of classical
evolving world of business, logistics and supply chain management true continuous criteria and Boolean ones from a methodological
are more and more important. point of view. Michel and Hentenryck (2004) present a very simple
Among supply chain studies, many papers on facility location tabu-search algorithm which performs amazingly well on the
problem have been published. Vlachopoulou, Silleos, and Manthou uncapacitated warehouse location problem. The algorithm uses a
(2001) aim at developing a geographic decision support system for linear neighborhood. Drezner, Scott, and Song (2003) concern with
the warehouse site selection process, enabling the manager to use the optimal location of a central warehouse, when the possible
quantitative and qualitative criteria in order to classify alternative locations and the number of warehouses are known. They solve
warehouses or visualize the best one. Sharma and Berry (2007) the problem sequentially. First, for any given central warehouse
consider the single stage capacitated warehouse location problem location, the problem is a pure inventory problem. They find the
(SSCWLP) where goods are shipped from plants to warehouses and optimal policy for the inventory problem. They express the total
from warehouses to markets. The problem is to choose a set of inventory and transportation costs as a function of the central
points where warehouses are located so that the sum of warehouse warehouse location. The next step is to optimize this total cost
location costs and transportation costs are minimized. In their function over all possible central warehouse locations. Partovi
study they consider different formulation styles due to Geoffrion (2006) explains a new analytic model for facility location that takes
and Graves (1974) and Sharma (1991) for the multistage ware- into account both external and internal criteria that sustain com-
house location problem; and cast them in the formulation style petitive advantage. Partovi’s model, which is based on quality func-
of Sharma and Sharma (2000) to obtain a variety of formulations tion deployment (QFD), also includes the analytic hierarchy
of the problem SSCWLP. A public warehouses selection support process (AHP) and the analytic network process (ANP) concepts
to determine the best location for a facility.
* Corresponding author.
The most well known general heuristic methods for facility
E-mail address: kahramanc@itu.edu.tr (C. Kahraman). location problems are Tabu Search (TS), Simulated Annealing

0957-4174/$ - see front matter Crown Copyright Ó 2009 Published by Elsevier Ltd. All rights reserved.
doi:10.1016/j.eswa.2009.11.022
3944 T. Demirel et al. / Expert Systems with Applications 37 (2010) 3943–3952

(SA), and Genetic Algorithms (GA). Arostegui, Kadipasaoglu, and science), industrial engineering, geography, economics, computer
Khumawala (2006) compare the relative performance of TS, SA, science, mathematics, marketing, electrical engineering, urban
and GA on various facilities location problems. Hidaka and Okano planning (Drezner, 1995). Various terms (e.g., facility location
(1997) propose a simulation-based approach to the large-scale selection, store choice, and site selection) are being used for loca-
incapacitated warehouse/facility location problems, including a tion selection, and they bear different meanings. For example, facil-
heuristic algorithm named ‘‘Balloon Search”. Many warehouse ity location problems are solved to minimize the total cost of
location models do not explicitly represent the effect of dependent serving all customers (e.g., a service center). Retail location prob-
criteria. Most of the traditional multi-criteria decision making lems, on the other hand, center on how to identify a location that
(MCDM) methods are based on the additive concept along with can maximize the number of shoppers (Cheng & Li, 2004).
the independence assumption, but each individual criterion is A rich source of seminal papers can be found from the field of
not always completely independent (Shee, Tzeng, & T, 2003). For location science, which regards location selection problems as geo-
solving the interactions among elements, the Analytic Network metrical and combinatorial optimization problems. Owen and Da-
Process (ANP) as a MCDM method was proposed by Saaty (1996). skin (1998) attributed a large amount of facility location selection
Kahraman, Ertay, and Büyüközkan (2006) develop fuzzy ANP and studies to operations research. A number of mathematical pro-
apply it to a company producing PVC window and door systems. gramming models have been developed, which can be classified
Warehouse location decisions are among the most critical deci- as (1) static and deterministic models, and (2) dynamic and stochastic
sions of distribution network design. The selection of a warehouse models. Current, Min, and Schilling (1990), on the other hand, re-
location among alternative locations is a multi-criteria decision- viewed that facility location are selected on the basis of four broad
making problem including both quantitative and qualitative crite- categories of objectives (i.e., cost minimization, demand-orienta-
ria. The conventional approaches to warehouse location selection tion, profit maximization, and environmental concerns). Static
problem tend to be less effective in dealing with the imprecise or and deterministic location mathematical models are the main-
vague nature of the linguistic assessment. Under many situations, stream research topics in this area, but which cannot address many
the values of the qualitative criteria are often imprecisely defined of the complicated location selection problems. In consideration of
for the decision-makers. changing demands over time, potential expansions, and future
Choquet integral has been used for the solution of multiple cri- relocations over the long term, dynamic and stochastic models
teria decision-making problems in the literature. Marichal, Meyer, are more robust in addressing these future uncertainty issues so
and Roubens (2005) analyze an ordinal sorting procedure (TOM- that the facilities built are able to remain operable over an ex-
ASO) for the assignment of alternatives to graded classes and pres- tended time period.
ent a freeware constructed from this procedure. Meyer and According to Owen and Daskin (1998), static and deterministic
Roubens (2006) present a multiple criteria decision support ap- models have five basic purposes, which focus on:
proach in order to build a ranking and suggest a best choice on a
set of alternatives. The aggregation is performed through the use  How to minimize the average distance traveled—This also corre-
of a fuzzy extension of the Choquet integral. This paper proposes sponds to minimization of travel time and traveling cost. For
a multi-criteria decision-making method using fuzzy integral for example, schools, retail shops, and emergency service centers
the evaluation of alternative warehouse locations. We first deter- must be near to target residents, and thus proximity is desirable
mine the main and sub-criteria and the hierarchy for the ware- for a wide range of public and private facilities.
house location selection problem, then make a multi-criteria  How to achieve maximum coverage—With respect to facilities
evaluation of the warehouse location alternatives to illustrate such as police stations, fire stations, community centers, and
how the generalized Choquet integral is used to do this. The Cho- hospitals, the most crucial is about how to determine the mini-
quet integral is a flexible aggregation operator being introduced mum number of facilities needed to cover all demands, given
by Sugeno (1974) and it is the generalization of the weighted aver- that they can serve within an acceptable time. Such a cost min-
age method, the Ordered Weighted Average (OWA) operator, and imization to save available resources is usually a policy for locat-
the max–min operator (Grabisch, Murofushi, & Sugeno, 2000). ing public facilities.
The Choquet integral is a generalization of the Lebesgue integral,  How to determine a center that can maximize coverage and mini-
defined with respect to a non classical measure, often called fuzzy mize the distance traveled. Owen and Daskin (1998) referred to
measure, or non-additive measure or also capacity. A fuzzy integral this as ‘‘(the minimization of) the maximum distance between
is a sort of very general averaging operator that can represent the any demand and its nearest facility”. This type of regret models
notions of importance of a criterion and interaction between considers opposing problems in terms of distance and coverage
criteria. in locating a center for building a facility.
The remainder of this paper is organized as follows. Section 2  How to deal with multiple objectives—Many location selection
presents multi-criteria methods for facility/warehouse location problems are inherently complex in nature, which involve mul-
selection problem. Section 3 introduces some definitions and for- tiple decision objectives. For example, locating a hospital may
mulations associated with Choquet integral. The steps of the meth- consider both location and allocation issues, while selecting a
odology used for warehouse location selection are given in Section site for a shopping mall may involve variables other than simply
4. Warehouse location criteria are defined in the Section 5. An minimal traveling cost and maximum coverage. Generally, most
application to a warehouse location selection problem in a Turkish of the objectives can be classified into one of the four general
logistic firm is presented in Section 6. The last section summarizes objective function categories suggested by Current et al.
the findings and makes suggestions for further research. (1990), which are cost, demand, profit, and environment.
 How to locate undesirable facility—Apart from the selection of
desirable facilities location, there are practical applications deal-
2. Multi-criteria methods for location selection ing with locating undesirable facilities that should be located
away from populations, such as waste disposal plants, airports,
The existing literature does not indicate a systematic fashion of water treatment centers, and nuclear plants.
location selection research. However, the study of location selec-
tion has a long and extensive history spanning many general re- In contrast to static and deterministic formulations, dynamic
search fields including operations research (or management and stochastic models are mainly dealing with planning for future
T. Demirel et al. / Expert Systems with Applications 37 (2010) 3943–3952 3945

conditions under two core uncertainty situations: (1) planning of individual criterion or their combination. In this paper, the gen-
with known model input parameters, and (2) planning with imper- eralized Choquet integral proposed by Auephanwiriyahul, Keller,
fect information of input parameters. They are described briefly as and Gader (2002) will be used, in which measurable evidence is
follows: represented in terms of intervals, whereas fuzzy measures are real
numbers, is an extension of the standard Choquet integral. In con-
1. Dynamic facility location models—These models usually attempt trast to Auephanwiriyahul et al. (2002) and Tsai and Lu (2006) pro-
to locate facilities over a specified time horizon by formulating poses another generalization that involves linguistic expressions as
such real-world problems as location–allocation, spatial, and well as information fusion between criteria to overcome vagueness
temporal aspects into either a single objective or multiple and imprecision of linguistic terms in questionnaires. Our method-
objectives with an optimal or near-optimal solution. These ology follows Tsai and Lu’s (2006) approach to Auephanwiriyahul
models assume that the future values of input parameters are et al. (2002).
known or vary deterministically over time. In the following, some definitions are given to explain the basics
2. Stochastic location models—These models assume that future of Choquet integral (Modave and Grabisch, 1998):
values of inputs are uncertain. It can be grouped into two Let I be the set of attributes (or any set in a general setting). A
approaches. Models that are developed under the probabilistic set function l : PðIÞ ! ½0; 1 is called a fuzzy measure if it satisfies
approach focus explicitly on the probability distributions of the three following axioms: (1) lðØÞ ¼ 0: an empty set has no
random variables, while the scenario approach embraces mod- importance, (2) lðIÞ ¼ 1: the maximal set has a maximal impor-
els that help generate a set of possible future values for the tance, (3) lðBÞ 6 lðCÞ if B, C  I and B  C: a new added criterion
variables. cannot make the importance of a coalition (a set of criteria) dimin-
ish. Therefore, in a problem where card(I)=n, a value for every ele-
As it is explained before, these methods do not take the interac- ment of P(I) including 2n values is needed. Assuming that the
tions among the location selection attributes into account and can values of the empty set and of the maximal set are fixed, ð2n  2Þ
not handle vague and incomplete information. The methods trying values or coefficients to define a fuzzy measure are needed. So,
to consider the interactions, like ANP, are very tedious since they there is clearly a trade-off between complexity and accuracy. How-
require a huge amount of calculations. The Choquet integral is an ever, the complexity can be significantly reduced in order to guar-
excellent multi-attribute tool for the problems having interactive antee that fuzzy measures are used in practical applications. A
attributes under fuzziness. fuzzy integral is a sort of weighted mean taking into account the
importance of every coalition of criteria.
The methodology is composed of eight steps (Tsai & Lu, 2006):
3. Choquet integral and the steps of the methodology
Step 1. Given criterion i, respondents’ linguistic preferences for
A fuzzy integral is a sort of general averaging operator that can the degree of importance, perceived performance levels
represent the notions of importance of a criterion and interactions of alternative locations, and tolerance zone are surveyed.
among criteria. To define fuzzy integrals, a set of values of impor- Step 2. In view of the compatibility between perceived perfor-
tance is needed. This set is composed of the values of a fuzzy mea- mance levels and the tolerance zone, trapezoidal fuzzy
sure. So, a value of importance for each subset of attributes is numbers are used to quantify all linguistic terms in this
needed. study. Given respondent t and criteria i, linguistic terms
The success of a Choquet integral depends on an appropriate for the degree of importance is parameterized by
representation of fuzzy measures, which captures the importance

Best location selection for warehouse location

Costs Labor Infrastructure Markets Macro


characteristics environment

• Labor costs • Skilled labor • Existence of modes • Proximity to • Policies of


• Transportation • Availability of of transportation customers government
costs labor force • Telecommunication • Proximity to • Industrial
• Tax incentives systems suppliers or regulations laws
and tax • Quality and producer • Zoning and
structures reliability of • Lead Times and construction plan
• Financial modes of responsiveness
incentives transportation
• Handling costs

Location A1 Location A2 Location A3 Location A4

Fig. 1. The hierarchy of the warehouse location selection problem.


3946 T. Demirel et al. / Expert Systems with Applications 37 (2010) 3943–3952

8 n 
e t ¼ ðat ; at ; at ; at Þ, perceived performance levels by e
A p ti ¼ >
> Q
i i1 i2 i3 i4 >
< 1=k ½1 þ kgðAi Þ  1 if k – 0
ðpti1 ; pti2 ; pti3 ; pti4 Þ, and the tolerance zone by e
e ti ¼ ðeti1L ; eti2L ; i¼1
1 ¼ gðSÞ ¼ ð6Þ
eti3U ; eti4U Þ. In this case study, t ¼ 1; 2; 3; 4; 5; i ¼ 1; 2; . . . ; >
> P
n
>
: gðAi Þ if k ¼ 0;
nj ; j ¼ 1; 2; 3; 4; n1 ¼ 3; n2 ¼ 2; n3 ¼ 4; n4 ¼ 3; where nj i¼1
represents the number of criteria in dimension j.
where, Ai \ Aj ¼ / for all i; j ¼ 1; 2; 3; . . . ; n and i–j, and
~t ; p
Step 3. Average A ~ti and ~ ~ i , and ~
eti into A ei , respectively using Eq. k 2 ð1; 1.
i
(1). Let l be a fuzzy measure on (I,P(I)) and an application
f : I ! Rþ . The Choquet integral off with respect to l
Pk ~ t Pk Pk Pk Pk !
t t t t is defined by:
~i ¼ t¼1 Ai t¼1 ai1 t¼1 ai2 t¼1 ai3 t¼1 ai4
A ¼ ; ; ; : Z
k k k k k X
n
ðCÞ fdl ¼ ðf ðrðiÞÞ  f ðrði  1ÞÞÞlðAðiÞ Þ ð7Þ
ð1Þ I i¼1

where r is a permutation of the indices in order to have


Step 4. Normalize the location value of each criterion using Eq.
(2). f ðrð1ÞÞ 6 . . . 6 f ðrðnÞÞ; AðiÞ ¼ frðiÞ; . . . ; rðnÞg and
~f ¼
i k f ai ¼ k ½fi;a ; fi;þa ; ð2Þ  f ðrð0ÞÞ ¼ 0; by convention:
a2½0;1 a2½0;1

where fi 2 FðSÞ is a fuzzy-valued function. e F ðSÞ is the set of It is easy to see that the Choquet integral is a Lebesgue integral
h i pa ea þ½1;1
all fuzzy-valued functions f ; fia ¼ fi;a ; fi;þa ¼ i i2 ai
;p up to a reordering of the indices. Actually, if the fuzzy measure l is
additive, then the Choquet integral reduces to a Lebesgue integral.
and eai are a-level cuts of e e i for all a ¼ ½0; 1.
p i and e It is shown in Modave & Grabisch (1998) that under rather gen-
eral assumptions over the set of alternatives X, and over the weak
Step 5. Find the location value of dimension j using Eq. (3). orders i , there exists a unique fuzzy measure l over I such that:
8x; y 2 X; x  y () uðxÞ P uðyÞ; ð8Þ
Z  Z Z 
ðCÞ ef d e
g¼ k ðCÞ  
fa dg a ; ðCÞ þ
fa dg a ;
þ
ð3Þ where
a¼½0;1
X
n
  
where gi : PðSÞ ! IðRþ Þ; gi ¼ ½g  þ a  þ  uðxÞ ¼ uðiÞ ðxðiÞ Þ  uði1Þ ðxði1Þ Þ l AðiÞ ; ð9Þ
i ; g i ; g i ¼ ½g i;a ; g i;a ; f i : S !
IðRþ Þ, and fi ¼ ½fi ; fiþ  for i ¼ 1; 2; 3; . . . ; nj : i¼1

To be able to calculate this location value, a k value and which is simply the aggregation of the monodimensional utility
the fuzzy measures gðAðiÞ Þ, i = 1, 2, . . . , n, are needed. These functions using the Choquet integral with respect to l.
are obtained from the following Eqs. (4)–(6) (Sugeno,
1974; Ishii & Sugeno, 1985): Step 6. Aggregate all dimensional performance levels of the loca-
gðAðnÞ Þ ¼ gðfsðnÞ gÞ ¼ g n ; ð4Þ tion alternatives into overall performance levels, using a
hierarchical process applying the two-stage aggregation
process of the generalized Choquet integral. This is repre-
gðAðiÞ Þ ¼ g i þ gðAðiþ1Þ Þ þ kg i gðAðiþ1Þ Þ; where 1 6 i < n
sented in Eq. (10). The overall performance levels yields a
ð5Þ fuzzy number, V e.

Fig. 2. Map of Turkey showing the alternative warehouse locations.


T. Demirel et al. / Expert Systems with Applications 37 (2010) 3943–3952 3947

R
maincriterionð1Þ ¼ ðCÞ fdg 4.2. Labor characteristics
Z
.. iV ¼ ðCÞ maincriterion dg ð10Þ
. This main criterion defines the state of qualified labor at a loca-
R
maincriterionðmÞ ¼ ðCÞ fdg tion and the degree of the availability of such labor. Skilled labor de-
fines the personnel for a work, who has qualities to perform such
e is l ~ ðxÞ; defuzzify the work and who is trained, qualified. This is one of the requirements
Step 7. Assume that the membership of V V
e into a crisp value v using Eq. (11) and in order to perform a work timely and in a qualified manner. The
fuzzy number V
skilled labor may not be at the desired level at each location. Avail-
make a comparison of the overall performance levels of
ability of labor force is a criterion that changes based on the level of
alternative locations.
development in the region, training levels, and population
a1 þ a2 þ a3 þ a4 structure.
e ¼
Fð AÞ : ð11Þ
4

Step 8. Compare weak and advantageous criteria among the loca- Table 2
The criteria of warehouse location selection and their symbols.
tions alternatives using Eq. (1).
Criteria The symbol of each
criterion

4. Warehouse selection criteria Costs C


1. Labor costs C1
2. Transportation costs C2
In this study, five main criteria and 16 sub-criteria of these main 3. Tax incentives and tax structures C3
criteria were used for the selection of warehouse location. These 4. Financial incentives C4
criteria were selected from the studies of Min & Melachrinoudis 5. Handling costs C5
(1999), Alberto (2000) and MacCarthy & Atthrirawong (2003). Labor characteristics LC
Moreover, while determining these criteria and their hierarchy, 1. Skilled labor LC 1
the opinions of experts in the logistics sector were also taken into 2. Availability of labor force LC 2
account. The definitions of the main and sub-criteria are summa- Infrastructure I
rized as follows: 1. Existence of modes of transportation I 1
2. Telecommunication systems I 2
3. Quality and reliability of modes of transportation I 3
4.1. Costs
Markets M
1. Proximity to customers M 1
Costs are one of the factors that the facility location highly af- 2. Proximity to suppliers or producer M 2
fects. Under the cost criterion, five sub-criteria are defined: labor 3. Lead Times and responsiveness M 3
costs, transportation cost, tax incentives and tax structures, financial Macro environment ME
incentives, and handling costs. Labor costs are the criterion that 1. Policies of government ME1
changes with respect to the life conditions at alternative locations. 2. Industrial regulations laws ME2
Transportation costs vary according to the economic structure of 3. Zoning and construction plan ME3

the alternative regions, transportation facilities and alternative


transportation types as air, land, railroad, and marine. Tax incen-
tives and tax structures vary based on the regions that have prior-
ity in development. The credit incentives by the state or private
Table 3
banks, which can be called Financial incentives, are realized in var-
Individual importance of criteria, the tolerance zones, and each warehouse location’s
ious regions. Handling costs, which is caused by the storage of the linguistic evaluation.
goods, are the costs that are composed of capital, work power,
Criteria Individual The Linguistic evaluation
equipment and risk costs and vary from a region to another
importance of tolerance
region. criteria zone
Izmir Izmit/ Zonguldak Bursa
Gebze
C HI
C1 M [M, H] SH M H H
Table 1 C2 HI [SL, SH] SH SH M SH
The relationship between trapezoidal fuzzy numbers and degrees of linguistic C3 SI [M, H] M M H M
importances in a nine-linguistic-term scale (Delgado et al., 1998). C4 SI [SL, SH] SH SH M SH
C5 HI [L, M] M M L SL
Low/high levels The degrees of Trapezoidal fuzzy
LC SU
importance numbers
LC 1 M [L, M] SL M L SL
Label Linguistic Label Linguistic terms LC 2 SU [SH, VH] SH VH SH H
terms
I HI
EL Extra low EU Extra (0, 0, 0, 0) I1 HI [M, VH] VH VH H H
unimportant I2 HI [H, EH] EH EH VH VH
VL Very low VU Very unimportant (0, 0.01, 0.02, 0.07) I3 VI [SH, VH] VH VH SH H
L Low U Unimportant (0.04, 0.1, 0.18, 0.23)
M VI
SL Slightly low SU Slightly (0.17, 0.22, 0.36, 0.42)
M1 SI [M, VH] M VH H H
unimportant
M2 SI [SH, VH] H VH H H
M Middle M Middle (0.32, 0.41, 0.58, 0.65)
M3 HI [SL, SH] M SH SL SH
SH Slightly high SI Slightly (0.58, 0.63, 0.8, 0.86)
important ME SI
H High HI High important (0.72, 0.78, 0.92, 0.97) ME1 HI [M, H] SH M H H
VH Very high VI Very important (0.93, 0.98, 0.98, 1.0) ME2 SI [SL, SH] SH SH M SH
EH Extra high EI Extra important (1, 1, 1, 1) ME3 SI [L, M] M M SL SL
3948 T. Demirel et al. / Expert Systems with Applications 37 (2010) 3943–3952

4.3. Infrastructure modes of transportation defines the transportation service between


the customer nodes, suppliers, and the warehouse, to be performed
It is a main criterion defining the basic structure of transporta- in a reliable and qualified way based on the different transporta-
tion, and communication systems, based on the location of a ware- tion modes. Reliable and quality service means timely delivery,
house. This criterion is composed of three sub-criteria: the delivery to the correct location and undamaged delivery of the
existence of modes of transportation, telecommunication, and goods.
quality and reliability of modes of transportation. Existence of
modes of transportation has an importance based on the availability 4.4. Market
of different transportation types in the location. Telecommunication
systems is a criterion that defines the communication facilities and It is a main criterion that defines the distance of the location
communication technologies of the warehouse with the customer of the warehouse to the customers, suppliers and producers.
nodes, the producers or the suppliers. Quality and reliability of Moreover, this criterion contains the supply periods and the abil-

Table 4
Compromised evaluations of five experts.

Criteria Individual importance The combined tolerance zone Perceived performance levels of alternative locations
_
Izmir _
Izmit-Gebze Zonguldak Bursa

C (0.72, 0.78,0.92, 0.97)


C1 (0.32, 0.41, 0.58, 0.65) (0.32, 0.41, 0.92, 0.97) (0.58, 0.63, 0.8, 0.86) (0.32, 0.41, 0.58, 0.65) (0.72, 0.78, 0.92, 0.97) (0.72, 0.78, 0.92, 0.97)
C2 (0.72, 0.78, 0.92, 0.97) (0.17, 0.22, 0.8, 0.86) (0.58, 0.63, 0.8, 0.86) (0.58, 0.63, 0.8, 0.86) (0.32, 0.41, 0.58, 0.65) (0.58, 0.63, 0.8, 0.86)
C3 (0.58, 0.63, 0.8, 0.86) (0.32, 0.41, 0.92, 0.97) (0.32, 0.41, 0.58, 0.65) (0.32, 0.41, 0.58, 0.65) (0.72, 0.78, 0.92, 0.97) (0.32, 0.41, 0.58, 0.65)
C4 (0.58, 0.63, 0.8, 0.86) (0.17, 0.22, 0.8, 0.86) (0.58, 0.63, 0.8, 0.86) (0.58, 0.63, 0.8, 0.86) (0.32, 0.41, 0.58, 0.65) (0.58, 0.63, 0.8, 0.86)
C5 (0.72, 0.78, 0.92, 0.97) (0.04, 0.1, 0.58, 0.65) (0.32, 0.41, 0.58, 0.65) (0.32, 0.41, 0.58, 0.65) (0.04, 0.1, 0.18, 0.23) (0.17, 0.22, 0.36, 0.42)
LC (0.17, 0.22, 0.36, 0.42)
LC 1 (0.32, 0.41, 0.58, 0.65) (0.04, 0.1, 0.58, 0.65) (0.17, 0.22, 0.36, 0.42) (0.32, 0.41, 0.58, 0.65) (0.04, 0.1, 0.18, 0.23) (0.17, 0.22, 0.36, 0.42)
LC 2 (0.17, 0.22, 0.36, 0.42) (0.58, 0.63, 0.98, 1.0) (0.58, 0.63, 0.8, 0.86) (0.93, 0.98, 0.98, 1.0) (0.58, 0.63, 0.8, 0.86) (0.72, 0.78, 0.92, 0.97)
I (0.72, 0.78, 0.92, 0.97)
I1 (0.72, 0.78, 0.92, 0.97) (0.32, 0.41, 0.98, 1.0) (0.93, 0.98, 0.98, 1.0) (0.93, 0.98, 0.98, 1.0) (0.72, 0.78, 0.92, 0.97) (0.72, 0.78, 0.92, 0.97)
I2 (0.72, 0.78, 0.92, 0.97) (0.72, 0.78, 1, 1) (1, 1, 1, 1) (1, 1, 1, 1) (0.93, 0.98, 0.98, 1.0) (0.93, 0.98, 0.98, 1.0)
I3 (0.93, 0.98, 0.98, 1.0) (0.58, 0.63, 0.98, 1.0) (0.93, 0.98, 0.98, 1.0) (0.93, 0.98, 0.98, 1.0) (0.58, 0.63, 0.8, 0.86) (0.72, 0.78, 0.92, 0.97)
M (0.93, 0.98, 0.98, 1.0)
M1 (0.58, 0.63, 0.8, 0.86) (0.32, 0.41, 0.98, 1.0) (0.32, 0.41, 0.58, 0.65) (0.93, 0.98, 0.98, 1.0) (0.72, 0.78, 0.92, 0.97) (0.72, 0.78, 0.92, 0.97)
M2 (0.58, 0.63, 0.8, 0.86) (0.58, 0.63, 0.98, 1.0) (0.72, 0.78, 0.92, 0.97) (0.93, 0.98, 0.98, 1.0) (0.72, 0.78, 0.92, 0.97) (0.72, 0.78, 0.92, 0.97)
M3 (0.72, 0.78, 0.92, 0.97) (0.17, 0.22, 0.8, 0.86) (0.32, 0.41, 0.58, 0.65) (0.58, 0.63, 0.8, 0.86) (0.17, 0.22, 0.36, 0.42) (0.58, 0.63, 0.8, 0.86)
ME (0.58, 0.63, 0.80, 0.86)
ME1 (0.72, 0.78, 0.92, 0.97) (0.32, 0.41, 0.92, 0.97) (0.58, 0.63, 0.8, 0.86) (0.32, 0.41, 0.58, 0.65) (0.72, 0.78, 0.92, 0.97) (0.72, 0.78, 0.92, 0.97)
ME2 (0.58, 0.63, 0.8, 0.86) (0.17, 0.22, 0.8, 0.86) (0.58, 0.63, 0.8, 0.86) (0.58, 0.63, 0.8, 0.86) (0.32, 0.41, 0.58, 0.65) (0.58, 0.63, 0.8, 0.86)
ME3 (0.58, 0.63, 0.8, 0.86) (0.04, 0.1, 0.58, 0.65) (0.32, 0.41, 0.58, 0.65) (0.32, 0.41, 0.58, 0.65) (0.17, 0.22, 0.36, 0.42) (0.17, 0.22, 0.36, 0.42)

Table 5
Evaluation results by the generalized Choquet integral for a ¼ 0.
R  R þ
Dimensions and criteria Individual importance of criteria The normalized discrepancy f i ¼ ½fi ; fiþ  and location value ½ðCÞ f  dg ; ðCÞ f þ dg 
þ
gi ¼ ½g 
i ; gi 
_
Izmir _
Izmit/Gebze Zonguldak Bursa

Costs [0.353, 0.8448] [0.3522, 0.8448] [0.3329, 0.8207] [0.3562, 0.8447]


C1 [0.32, 0.65] [0.305, 0.77] [0.175, 0.665] [0.375, 0.825] [0.375, 0.825]
C2 [0.72, 0.97] [0.36, 0.845] [0.36, 0.845] [0.23, 0.74] [0.36, 0.845]
C3 [0.58, 0.86] [0.175, 0.665] [0.175, 0.665] [0.375, 0.825] [0.175, 0.665]
C4 [0.58, 0.86] [0.36, 0.845] [0.36, 0.845] [0.23, 0.74] [0.36, 0.845]
C5 [0.72, 0.97] [0.335, 0.805] [0.335, 0.805] [0.195, 0.595] [0.26, 0.69]
Labor characteristics [0.265, 0.625] [0.357, 0.771] [0.211, 0.6139] [0.277, 0.692]
LC 1 [0.32, 0.65] [0.26, 0.69] [0.335, 0.805] [0.195, 0.595] [0.26, 0.69]
LC 2 [0.17, 0.42] [0.29, 0.64] [0.465, 0.71] [0.29, 0.64] [0.36, 0.695]
Infrastructure [0.49, 0.8361] [0.49, 0.8361] [0.43, 0.8194] [0.4356, 0.8211]
I1 [0.72, 0.97] [0.465, 0.84] [0.465, 0.84] [0.36, 0.825] [0.36, 0.825]
I2 [0.72, 0.97] [0.5, 0.64] [0.5, 0.64] [0.465, 0.64] [0.465, 0.64]
I3 [0.93, 1.0] [0.465, 0.71] [0.465, 0.71] [0.29, 0.64] [0.36, 0.695]
Markets [0.2992, 0.7385] [0.4489, 0.8443] [0.3286, 0.8054] [0.36, 0.8438]
M1 [0.58, 0.86] [0.16, 0.665] [0.465, 0.84] [0.36, 0.825] [0.36, 0.825]
M2 [0.58, 0.86] [0.36, 0.695] [0.465, 0.71] [0.36, 0.695] [0.36, 0.695]
M3 [0.72, 0.97] [0.23, 0.74] [0.36, 0.845] [0.155, 0.625] [0.36, 0.845]
Macro environment [0.3449, 0.8387] [0.325, 0.8366] [0.34, 0.8222] [0.3619, 0.8416]
ME1 [0.72, 0.97] [0.305, 0.77] [0.175, 0.665] [0.375, 0.825] [0.375, 0.825]
ME2 [0.58, 0.86] [0.36, 0.845] [0.36, 0.845] [0.23, 0.74] [0.36, 0.845]
ME3 [0.58, 0.86] [0.335, 0.805] [0.335, 0.805] [0.26, 0.69] [0.26, 0.69]
T. Demirel et al. / Expert Systems with Applications 37 (2010) 3943–3952 3949

ity to respond to an order. Proximity to customers defines the dis- rion that differs from a region to another region. Being parallel to
tance of the warehouse location to the customer nodes. Proximity the development level of the region, it includes the factors such
to suppliers or producers defines the distance of the warehouse as various incentives, tax exemptions, and investment facilities.
location to the suppliers and the producers. Lead times and Industrial regulations laws, which is parallel to the policies and
responsiveness defines the ability and the period to fulfill an implementations of the local administrations, is a criterion that de-
order. fines various laws and arrangements. Zoning and construction plan
is a sub-criterion that defines the different development plans,
4.5. Macro environment implementations and arrangements of local administrations at
alternative locations.
This main criterion includes the government policies, industrial Employing the main and sub-criteria explained above, a hierar-
arrangements and laws, and the development plans of the region at chical structure for a warehouse location selection problem is
the macro level. Policies of government is considered as a sub-crite- shown in Fig. 1.

Table 6
Fuzzy measures.

_
Izmir _
Izmit/Gebze Zonguldak Bursa

g  ¼ ðAðiÞ Þ g þ ¼ ðAðiÞ Þ g  ¼ ðAðiÞ Þ g þ ¼ ðAðiÞ Þ g  ¼ ðAðiÞ Þ g þ ¼ ðAðiÞ Þ g  ¼ ðAðiÞ Þ g þ ¼ ðAðiÞ Þ

Fuzzy measures
k ¼ 0:989 k ¼ 0:999 k ¼ 0:989 k ¼ 0:999 k ¼ 0:989 k ¼ 0:999 k ¼ 0:989 k ¼ 0:999
g  ðAð2Þ Þ ¼ 0:98 g þ ðAð2Þ Þ ¼ 1 g  ðAð1Þ Þ ¼ 1 g þ ðAð1Þ Þ ¼ 1 g  ðAð4Þ Þ ¼ 0:716 g þ ðAð4Þ Þ ¼ 0:95 g  ðAð5Þ Þ ¼ 0:32 g þ ðAð3Þ Þ ¼ 0:999
g  ðAð4Þ Þ ¼ 0:88 g þ ðAð4Þ Þ ¼ 0:9966 g  ðAð4Þ Þ ¼ 0:88 g þ ðAð4Þ Þ ¼ 0:9966 g  ðAð2Þ Þ ¼ 0:975 g þ ðAð2Þ Þ ¼ 1 g  ðAð3Þ Þ ¼ 0:926 g þ ðAð4Þ Þ ¼ 0:9966
g  ðAð1Þ Þ ¼ 1 g þ ðAð1Þ Þ ¼ 1 g  ðAð2Þ Þ ¼ 0:99 g þ ðAð2Þ Þ ¼ 1 g  ðAð5Þ Þ ¼ 0:58 g þ ðAð5Þ Þ ¼ 0:86 g  ðAð1Þ Þ ¼ 1 g þ ðAð1Þ Þ ¼ 1
g  ðAð5Þ Þ ¼ 0:58 g þ ðAð5Þ Þ ¼ 0:86 g  ðAð5Þ Þ ¼ 0:58 g þ ðAð5Þ Þ ¼ 0:86 g  ðAð3Þ Þ ¼ 0:885 g þ ðAð3Þ Þ ¼ 0:994 g  ðAð4Þ Þ ¼ 0:716 g þ ðAð3Þ Þ ¼ 0:86
g  ðAð3Þ Þ ¼ 0:97 g þ ðAð3Þ Þ ¼ 1 g  ðAð3Þ Þ ¼ 0:97 g þ ðAð3Þ Þ ¼ 1 g  ðAð1Þ Þ ¼ 1 g þ ðAð1Þ Þ ¼ 1 g  ðAð2Þ Þ ¼ 0:9867 g þ ðAð2Þ Þ ¼ 1

k ¼ 9:373 k ¼ 0:2567 k ¼ 9:373 k ¼ 0:2567 k ¼ 9:373 k ¼ 0:2567 k ¼ 9:373 k ¼ 0:2567


g  ðAð1Þ Þ ¼ 1 g þ ðAð2Þ Þ ¼ 0:65 g  ðAð1Þ Þ ¼ 1 g þ ðAð2Þ Þ ¼ 0:65 g  ðAð1Þ Þ ¼ 1 g þ ðAð1Þ Þ ¼ 1 g  ðAð1Þ Þ ¼ 1 g þ ðAð1Þ Þ ¼ 1
g  ðAð2Þ Þ ¼ 0:17 g þ ðAð1Þ Þ ¼ 1 g  ðAð2Þ Þ ¼ 0:17 g þ ðAð1Þ Þ ¼ 1 g  ðAð2Þ Þ ¼ 0:17 g þ ðAð2Þ Þ ¼ 0:42 g  ðAð2Þ Þ ¼ 0:17 g þ ðAð2Þ Þ ¼ 0:42

k ¼ 0:9939 k ¼ 0:9999 k ¼ 0:9939 k ¼ 0:9999 k ¼ 0:9939 k ¼ 0:9999 k ¼ 0:9939 k ¼ 0:9999


g  ðAð1Þ Þ ¼ 1 g þ ðAð3Þ Þ ¼ 0:97 g  ðAð1Þ Þ ¼ 1 g þ ðAð3Þ Þ ¼ 0:97 g  ðAð2Þ Þ ¼ 0:9247 g þ ðAð3Þ Þ ¼ 0:97 g  ðAð1Þ Þ ¼ 1 g þ ðAð3Þ Þ ¼ 0:97
g  ðAð3Þ Þ ¼ 0:72 g þ ðAð1Þ Þ ¼ 1 g  ðAð3Þ Þ ¼ 0:72 g þ ðAð1Þ Þ ¼ 1 g  ðAð3Þ Þ ¼ 0:72 g þ ðAð1Þ Þ ¼ 1 g  ðAð3Þ Þ ¼ 0:72 g þ ðAð1Þ Þ ¼ 1
g  ðAð2Þ Þ ¼ 0:9845 g þ ðAð2Þ Þ ¼ 1 g  ðAð2Þ Þ ¼ 0:9845 g þ ðAð2Þ Þ ¼ 1 g  ðAð1Þ Þ ¼ 1 g þ ðAð2Þ Þ ¼ 1 g  ðAð2Þ Þ ¼ 0:9845 g þ ðAð2Þ Þ ¼ 1

k ¼ 0:9299 k ¼ 0:9994 k ¼ 0:9299 k ¼ 0:9994 k ¼ 0:9299 k ¼ 0:9994 k ¼ 0:9299 k ¼ 0:9994


g  ðAð1Þ Þ ¼ 1 g þ ðAð1Þ Þ ¼ 1 g  ðAð2Þ Þ ¼ 0:847 g þ ðAð2Þ Þ ¼ 0:996 g  ðAð2Þ Þ ¼ 0:847 g þ ðAð3Þ Þ ¼ 0:86 g  ðAð1Þ Þ ¼ 1 g þ ðAð2Þ Þ ¼ 0:996
g  ðAð3Þ Þ ¼ 0:58 g þ ðAð2Þ Þ ¼ 0:996 g  ðAð3Þ Þ ¼ 0:58 g þ ðAð1Þ Þ ¼ 1 g  ðAð3Þ Þ ¼ 0:58 g þ ðAð2Þ Þ ¼ 0:98 g  ðAð2Þ Þ ¼ 0:9117 g þ ðAð1Þ Þ ¼ 1
g  ðAð2Þ Þ ¼ 0:9117 g þ ðAð3Þ Þ ¼ 0:97 g  ðAð1Þ Þ ¼ 1 g þ ðAð3Þ Þ ¼ 0:97 g  ðAð1Þ Þ ¼ 1 g þ ðAð1Þ Þ ¼ 1 g  ðAð3Þ Þ ¼ 0:72 g þ ðAð3Þ Þ ¼ 0:97

k ¼ 0:9299 k ¼ 0:9994 k ¼ 0:9299 k ¼ 0:9994 k ¼ 0:9299 k ¼ 0:9994 k ¼ 0:9299 k ¼ 0:9994


g  ðAð1Þ Þ ¼ 1 g þ ðAð1Þ Þ ¼ 1 g  ðAð1Þ Þ ¼ 1 g þ ðAð1Þ Þ ¼ 1 g  ðAð3Þ Þ ¼ 0:72 g þ ðAð3Þ Þ ¼ 0:97 g  ðAð3Þ Þ ¼ 0:72 g þ ðAð2Þ Þ ¼ 0:996
g  ðAð3Þ Þ ¼ 0:58 g þ ðAð3Þ Þ ¼ 0:86 g  ðAð3Þ Þ ¼ 0:58 g þ ðAð3Þ Þ ¼ 0:86 g  ðAð1Þ Þ ¼ 1 g þ ðAð2Þ Þ ¼ 0:996 g  ðAð2Þ Þ ¼ 0:9117 g þ ðAð3Þ Þ ¼ 0:86
g  ðAð2Þ Þ ¼ 0:847 g þ ðAð2Þ Þ ¼ 0:98 g  ðAð2Þ Þ ¼ 0:847 g þ ðAð2Þ Þ ¼ 0:98 g  ðAð2Þ Þ ¼ 0:9117 g þ ðAð1Þ Þ ¼ 1 g  ðAð1Þ Þ ¼ 1 g þ ðAð1Þ Þ ¼ 1

Table 7
Evaluation results by the generalized Choquet integral for a ¼ 1.
R  R þ
Dimensions and criteria Individual importance of criteria The normalized discrepancy f i ¼ ½fi ; fiþ  and location value ½ðCÞ f  dg ; ðCÞ f þ dg 
þ
gi ¼ ½g 
i ; gi 
_
Izmir _
Izmit/Gebze Zonguldak Bursa

Costs [0.4119, 0.789] [0.4125, 0.789] [0.4021, 0.7485] [0.4163, 0.7885]


C1 [0.41, 0.58] [0.355, 0.695] [0.245, 0.585] [0.43, 0.755] [0.43, 0.755]
C2 [0.78, 0.92] [0.415, 0.79] [0.415, 0.79] [0.305, 0.68] [0.415, 0.79]
C3 [0.63, 0.8] [0.245, 0.585] [0.245, 0.585] [0.43, 0.755] [0.245, 0.585]
C4 [0.63, 0.8] [0.415, 0.79] [0.415, 0.79] [0.305, 0.68] [0.415, 0.79]
C5 [0.78,0.92] [0.415, 0.74] [0.415, 0.74] [0.26, 0.54] [0.32, 0.63]
Labor characteristics [0.3211, 0.6111] [0.4337, 0.7127] [0.2743, 0.5562] [0.3376, 0.6354]
LC 1 [0.41, 0.58] [0.32, 0.63] [0.415, 0.74] [0.26, 0.54] [0.32, 0.63]
LC 2 [0.22, 0.36] [0.325, 0.585] [0.5, 0.675] [0.325, 0.585] [0.4, 0.645]
Infrastructure [0.5, 0.7824] [0.5, 0.7824] [0.46, 0.7423] [0.4648, 0.7461]
I1 [0.78, 0.92] [0.5, 0.785] [0.5, 0.785] [0.4, 0.755] [0.4, 0.755]
I2 [0.72, 0.78] [0.5, 0.61] [0.5, 0.61] [0.49, 0.6] [0.49, 0.6]
I3 [0.98, 0.98] [0.5, 0.675] [0.5, 0.675] [0.325, 0.585] [0.4, 0.645]
Markets [0.3592, 0.6763] [0.4896, 0.7881] [0.3769, 0.7301] [0.4117, 0.7857]
M1 [0.63, 0.8] [0.215, 0.585] [0.5, 0.785] [0.4, 0.755] [0.4, 0.755]
M2 [0.63, 0.8] [0.4, 0.645] [0.5, 0.675] [0.4, 0.645] [0.4, 0.645]
M3 [0.78, 0.92] [0.305, 0.68] [0.415, 0.79] [0.21, 0.57] [0.415, 0.79]
Macro environment [0.4077, 0.7782] [0.3943, 0.7741] [0.4048, 0.768] [0.42, 0.7813]
ME1 [0.78,0.92] [0.355, 0.695] [0.245, 0.585] [0.43, 0.755] [0.43, 0.755]
ME2 [0.63, 0.8] [0.415, 0.79] [0.415, 0.79] [0.305, 0.68] [0.415, 0.79]
ME3 [0.63, 0.8] [0.415, 0.74] [0.415, 0.74] [0.32, 0.63] [0.32, 0.63]
3950 T. Demirel et al. / Expert Systems with Applications 37 (2010) 3943–3952

5. Application to a warehouse location selection problem in a The other normalized discrepancies between for location 1 and
Turkish logistic firm main criterion C at /¼ 0 are f 02 ¼ ½0:36; 0:845; f 03 ¼ ½0:175; 0:665;
f 0 ¼ ½0:36; 0:845; and f 0 ¼ ½0:335; 0:805, respectively. Their corre-
4 5
Anadolum Logistics, a big logistics company in Turkey, wants to sponding degrees of individual importance are g01 ¼ ½0:32; 0:65;
decide on where it will locate its new warehouse. Since we were the 02 ¼ ½0:72; 0:97; g03 ¼ ½0:58; 0:86; g04 ¼ ½0:58; 0:86,
g and g05 ¼

managerial consultants of this company, we presented the criteria ½0:72; 0:97, respectively. First, the sequence fi;0 is sorted, where
to the experts’ team, given above in Fig. 1. The alternative locations i = 1, 2, 3, 4 and 5, as follows:
have been determined by the five experts of the company: Izmir,
_ fC3 ¼ 0:175 < fC1 ¼ 0:305< fC5 ¼ 0:335< fC2 ¼ 0:36< fC4 ¼ 0:36
Bursa, Izmit/Gebze, and Zonguldak. These alternative locations
g
C 3 ¼ 0:58 g
C 1 ¼ 0:32 g
C 5 ¼ 0:72 g
C 2 ¼ 0:72 g
C 4 ¼ 0:58
are shown in Fig. 2. The experts were from the departments Logistics
g 1 ¼ 0:58 g 2 ¼ 0:32 g 3 ¼ 0:72 g 4 ¼ 0:72 g 5 ¼ 0:58
(3 experts), Marketing (1 expert), and Finance (1 expert). These five
experts confirmed the criteria and sub-criteria and decided on
By solving the following equation for k, the fuzzy measures
using the evaluation scale in Table 1. To express the sub-criteria
gðAðiÞ Þ,i=1,2,...,n are obtained as follows:
easier, the symbols in Table 2 were generated. The team generated
the individual importance of main and sub-criteria and their toler- 1
1 ¼ gðSÞ ¼ f½ð1 þ k0:58Þð1 þ k0:32Þð1 þ k0:72Þ
ance zones and made each warehouse location’s linguistic evalua- k
tion. Table 3 gives these results.  ð1 þ k0:72Þð1 þ k0:58Þ  1g:
Trapezoidal fuzzy numbers are used to quantify the linguistic
That is,
terms in Table 3. The tolerance zones in this table are obtained
in that way: the first two numerical values of the lower linguistic k ¼ 0:9898:
value of a tolerance zone in Table 3 are combined with the last
The fuzzy measures are,
two numerical values of the upper linguistic value of the same tol-
erance zone. Consider the tolerance zone [M, H]. The corresponding gðAð 5ÞÞ ¼ g 5 ¼ 0:58;
numerical values of M and H are (0.32, 0.41, 0.58, 0.65) and gðAð4Þ Þ ¼ g 4 þ gðAð5Þ Þ þ kg 4 gðAð5Þ Þ ¼ 0:88;
(0.72, 0.78, 0.92, 0.97), respectively. Then the combined tolerance
gðAð3Þ Þ ¼ g 3 þ gðAð4Þ Þ þ kg 3 gðAð4Þ Þ ¼ 0:97;
zone is (0.32, 0.41, 0.92, 0.97). Table 4 presents the compromised
evaluations of the experts’ team. If it is requested, the separate gðAð2Þ Þ ¼ g 2 þ gðAð3Þ Þ þ kg 2 gðAð3Þ Þ ¼ 0:98;
evaluations of each expert can be combined by calculating the gðAð1Þ Þ ¼ g 1 þ gðAð2Þ Þ þ kg 1 gðAð2Þ Þ ¼ 1:0:
arithmetic mean of these numerical values.
Tables 6 and 8 summarize the whole fuzzy measures and k values,
Table 5 gives the evaluation results by the generalized Choquet
which are calculated in the same way above.
Integral for a ¼ 0. For the sub-criteria, Eq. (2) is used while Eq. (3)
The aggregated Choquet integral values for the main criterion C
is for the main criteria. For example, the value [0.305, 0.77] of
are calculated as in the following. Tables 5 and 7 include the nor-
‘‘location 1 and sub-criterion C 1 ” is obtained in that way:
malized discrepancies and location values (Choquet integrals).
Z
½0:58; 0:86  ½0:32; 0:97 þ ½1; 1
f ; fia ¼ ½fi;a ; fi;þa  ¼ ¼ ½0:305; 0:77: ðCÞ

fa¼0 dg a¼0 ¼ 1ð0:175Þ þ 0:98ð0:305  0:175Þ
2
þ 0:97ð0:335  0:305Þ þ 0:88ð0:36  0:335Þ
For the value [0.353, 0.8448] of ‘‘location 1 and main-criterion C” is þ 0:58ð0:36  0:36Þ ¼ 0:353:
obtained in that way:

Table 8
Fuzzy measures.

_
Izmir _
Izmit/Gebze Zonguldak Bursa

g  ¼ ðAðiÞ Þ g þ ¼ ðAðiÞ Þ g  ¼ ðAðiÞ Þ g þ ¼ ðAðiÞ Þ g  ¼ ðAðiÞ Þ g þ ¼ ðAðiÞ Þ g  ¼ ðAðiÞ Þ g þ ¼ ðAðiÞ Þ

Fuzzy measures
k ¼ 0:996 k ¼ 0:9999 k ¼ 0:996 k ¼ 0:9999 k ¼ 0:996 k ¼ 0:9999 k ¼ 0:996 k ¼ 0:9999
g  ðAð2Þ Þ ¼ 0:986 g þ ðAð2Þ Þ ¼ 0:9995 g  ðAð1Þ Þ ¼ 1 g þ ðAð1Þ Þ ¼ 1 g  ðAð4Þ Þ ¼ 0:7827 g þ ðAð4Þ Þ ¼ 0:916 g  ðAð5Þ Þ ¼ 0:41 g þ ðAð3Þ Þ ¼ 0:9933
g  ðAð3Þ Þ ¼ 0:974 g þ ðAð4Þ Þ ¼ 0:984 g  ðAð3Þ Þ ¼ 0:9853 g þ ðAð4Þ Þ ¼ 0:984 g  ðAð2Þ Þ ¼ 0:9856 g þ ðAð2Þ Þ ¼ 0:9987 g  ðAð3Þ Þ ¼ 0:9546 g þ ðAð4Þ Þ ¼ 0:984
g  ðAð1Þ Þ ¼ 1 g þ ðAð1Þ Þ ¼ 1 g  ðAð2Þ Þ ¼ 0:997 g þ ðAð2Þ Þ ¼ 0:9998 g  ðAð5Þ Þ ¼ 0:63 g þ ðAð5Þ Þ ¼ 0:80 g  ðAð1Þ Þ ¼ 1 g þ ðAð1Þ Þ ¼ 1
g  ðAð4Þ Þ ¼ 0:87 g þ ðAð5Þ Þ ¼ 0:80 g  ðAð4Þ Þ ¼ 0:9205 g þ ðAð5Þ Þ ¼ 0:80 g  ðAð3Þ Þ ¼ 0:9215 g þ ðAð3Þ Þ ¼ 0:9832 g  ðAð4Þ Þ ¼ 0:7827 g þ ðAð5Þ Þ ¼ 0:80
g  ðAð5Þ Þ ¼ 0:78 g þ ðAð3Þ Þ ¼ 0:9988 g  ðAð5Þ Þ ¼ 0:78 g þ ðAð3Þ Þ ¼ 0:9988 g  ðAð1Þ Þ ¼ 1 g þ ðAð1Þ Þ ¼ 1 g  ðAð2Þ Þ ¼ 0:9929 g þ ðAð2Þ Þ ¼ 0:9995

k ¼ 4:1 k ¼ 0:2868 k ¼ 4:1 k ¼ 0:2868 k ¼ 4:1 k ¼ 0:2868 k ¼ 4:1 k ¼ 0:2868


g  ðAð1Þ Þ ¼ 1 g þ ðAð2Þ Þ ¼ 0:58 g  ðAð1Þ Þ ¼ 1 g þ ðAð2Þ Þ ¼ 0:58 g  ðAð1Þ Þ ¼ 1 g þ ðAð1Þ Þ ¼ 1 g  ðAð1Þ Þ ¼ 1 g þ ðAð1Þ Þ ¼ 1
g  ðAð2Þ Þ ¼ 0:22 g þ ðAð1Þ Þ ¼ 1 g  ðAð2Þ Þ ¼ 0:22 g þ ðAð1Þ Þ ¼ 1 g  ðAð2Þ Þ ¼ 0:22 g þ ðAð2Þ Þ ¼ 0:36 g  ðAð2Þ Þ ¼ 0:22 g þ ðAð2Þ Þ ¼ 0:36

k ¼ 0:9987 k ¼ 0:9997 k ¼ 0:9987 k ¼ 0:9997 k ¼ 0:9987 k ¼ 0:9997 k ¼ 0:9987 k ¼ 0:9997


g  ðAð1Þ Þ ¼ 1 g þ ðAð3Þ Þ ¼ 0:92 g  ðAð1Þ Þ ¼ 1 g þ ðAð3Þ Þ ¼ 0:92 g  ðAð2Þ Þ ¼ 0:9391 g þ ðAð3Þ Þ ¼ 0:92 g  ðAð1Þ Þ ¼ 1 g þ ðAð3Þ Þ ¼ 0:92
g  ðAð2Þ Þ ¼ 0:9953 g þ ðAð1Þ Þ ¼ 1 g  ðAð2Þ Þ ¼ 0:9953 g þ ðAð1Þ Þ ¼ 1 g  ðAð3Þ Þ ¼ 0:72 g þ ðAð2Þ Þ ¼ 0:9826 g  ðAð3Þ Þ ¼ 0:72 g þ ðAð1Þ Þ ¼ 1
g  ðAð3Þ Þ ¼ 0:98 g þ ðAð2Þ Þ ¼ 0:9986 g  ðAð3Þ Þ ¼ 0:98 g þ ðAð2Þ Þ ¼ 0:9986 g  ðAð1Þ Þ ¼ 1 g þ ðAð1Þ Þ ¼ 1 g  ðAð2Þ Þ ¼ 0:9953 g þ ðAð2Þ Þ ¼ 0:9986

k ¼ 0:9611 k ¼ 0:9966 k ¼ 0:9611 k ¼ 0:9966 k ¼ 0:9611 k ¼ 0:9966 k ¼ 0:9611 k ¼ 0:9966


g  ðAð1Þ Þ ¼ 1 g þ ðAð1Þ Þ ¼ 1 g  ðAð2Þ Þ ¼ 0:8785 g þ ðAð2Þ Þ ¼ 0:9865 g  ðAð2Þ Þ ¼ 0:8785 g þ ðAð3Þ Þ ¼ 0:80 g  ðAð1Þ Þ ¼ 1 g þ ðAð2Þ Þ ¼ 0:9865
g  ðAð3Þ Þ ¼ 0:63 g þ ðAð2Þ Þ ¼ 0:9865 g  ðAð3Þ Þ ¼ 0:63 g þ ðAð1Þ Þ ¼ 1 g  ðAð3Þ Þ ¼ 0:63 g þ ðAð2Þ Þ ¼ 0:9621 g  ðAð2Þ Þ ¼ 0:9377 g þ ðAð1Þ Þ ¼ 1
g  ðAð2Þ Þ ¼ 0:9377 g þ ðAð3Þ Þ ¼ 0:92 g  ðAð1Þ Þ ¼ 1 g þ ðAð3Þ Þ ¼ 0:92 g  ðAð1Þ Þ ¼ 1 g þ ðAð1Þ Þ ¼ 1 g  ðAð3Þ Þ ¼ 0:78 g þ ðAð3Þ Þ ¼ 0:92

k ¼ 0:9611 k ¼ 0:9966 k ¼ 0:9611 k ¼ 0:9966 k ¼ 0:9611 k ¼ 0:9966 k ¼ 0:9611 k ¼ 0:9966


g  ðAð1Þ Þ ¼ 1 g þ ðAð1Þ Þ ¼ 1 g  ðAð1Þ Þ ¼ 1 g þ ðAð1Þ Þ ¼ 1 g  ðAð3Þ Þ ¼ 0:78 g þ ðAð3Þ Þ ¼ 0:92 g  ðAð3Þ Þ ¼ 0:78 g þ ðAð2Þ Þ ¼ 0:9865
g  ðAð2Þ Þ ¼ 0:8785 g þ ðAð3Þ Þ ¼ 0:80 g  ðAð2Þ Þ ¼ 0:8785 g þ ðAð3Þ Þ ¼ 0:80 g  ðAð1Þ Þ ¼ 1 g þ ðAð2Þ Þ ¼ 0:9865 g  ðAð2Þ Þ ¼ 0:9377 g þ ðAð3Þ Þ ¼ 0:80
g  ðAð3Þ Þ ¼ 0:63 g þ ðAð2Þ Þ ¼ 0:9621 g  ðAð3Þ Þ ¼ 0:63 g þ ðAð2Þ Þ ¼ 0:9621 g  ðAð2Þ Þ ¼ 0:9377 g þ ðAð1Þ Þ ¼ 1 g  ðAð1Þ Þ ¼ 1 g þ ðAð1Þ Þ ¼ 1
Table 9
Defuzzified overall values of alternative locations using generalized Chouqet integral.
R R
Criteria ðCÞ ~f dg~ Defuzzified ðCÞ ~f dg~

T. Demirel et al. / Expert Systems with Applications 37 (2010) 3943–3952


Izmir Izmit/Gebze Zonguldak Bursa _
Izmir _
Izmit/Gebze Zonguldak Bursa

Overall location value (0.449, 0.479, 0.788, 0.844) (0.485, 0.497, 0.789, 0.844) (0.404, 0.447, 0.748, 0.822) (0.414, 0.456, 0.788, 0.844) 0.640 0.653* 0.605 0.625
Costs (0.353, 0.4119, 0.789, 0.8448) (0.3522, 0.4125, 0.789, 0.8448) (0.3329, 0.4021, 0.7485, 0.8207) (0.3562, 0.4163, 0.7885, 0.8447) 0.599 0.599 0.576 0.601*
C1 (0.305, 0.355, 0.695, 0.77) (0.175, 0.245,0.585, 0.665) (0.375, 0.43, 0.755, 0.825) (0.375, 0.43, 0.755, 0.825) 0.531 0.417 0.596* 0.596*
C2 (0.36, 0.415, 0.79, 0.845) (0.36, 0.415, 0.79, 0.845) (0.23, 0.305, 0.68, 0.74) (0.36, 0.415, 0.79, 0.845) 0.602* 0.602* 0.489 0.602*
C3 (0.175, 0.245,0.585, 0.665) (0.175, 0.245,0.585, 0.665) (0.375, 0.43, 0.755, 0.825) (0.175, 0.245,0.585, 0.665) 0.417 0.417 0.596* 0.417
C4 (0.36, 0.415, 0.79, 0.845) (0.36, 0.415, 0.79, 0.845) (0.23, 0.305, 0.68, 0.74) (0.36, 0.415, 0.79, 0.845) 0.602* 0.602* 0.489 0.602*
C5 (0.335, 0.415, 0.74, 0.805) (0.335, 0.415, 0.74, 0.805) (0.195, 0.26, 0.54, 0.595) (0.26, 0.32, 0.63, 0.69) 0.574* 0.574* 0.397 0.475
Labor characteristics (0.265, 0.3211, 0.6111, 0.625) (0.357, 0.4337, 0.7127, 0.771) (0.211, 0.2743, 0.5562, 0.6139) (0.277, 0.3376, 0.6354, 0.692) 0.456 0.569* 0.414 0.486
LC 1 (0.26, 0.32, 0.63, 0.69) (0.335, 0.415, 0.74, 0.805) (0.195, 0.26, 0.54, 0.595) (0.26, 0.32, 0.63, 0.69) 0.475 0.574* 0.397 0.475
LC 2 (0.29, 0.325, 0.585, 0.64) (0.465, 0.5, 0.675, 0.71) (0.29, 0.325, 0.585, 0.64) (0.36, 0.4, 0.645, 0.695) 0.460 0.587* 0.460 0.525
Infrastructure (0.49, 0.5, 0.7824, 0.8361) (0.49, 0.5, 0.7824, 0.8361) (0.43, 0.46, 0.7423, 0.8194) (0.4356, 0.4648, 0.7461, 0.8211) 0.652* 0.652* 0.613 0.617
I1 (0.465, 0.5, 0.785, 0.84) (0.465, 0.5, 0.785, 0.84) (0.36, 0.4, 0.755, 0.825) (0.36, 0.4, 0.755, 0.825) 0.647* 0.647* 0.585 0.585
I2 (0.5, 0.5, 0.61, 0.64) (0.5, 0.5, 0.61, 0.64) (0.465, 0.49,0.6,0.64) (0.465, 0.49, 0.6, 0.64) 0.562* 0.562* 0.549 0.549
I3 (0.465, 0.5, 0.675, 0.71) (0.465, 0.5,0.675, 0.71) (0.29, 0.325, 0.585, 0.64) (0.36, 0.4, 0.645, 0.695) 0.587* 0.587* 0.460 0.525
Markets (0.2992, 0.3592, 0.6763, 0.7385) (0.4489, 0.4896, 0.7881, 0.8443) (0.3286, 0.3769, 0.7301, ,0.8054) (0.36, 0.4117, 0.7857, 0.8438) 0.518 0.642* 0.560 0.600
M1 (0.16, 0.215, 0.585, 0.665) (0.465, 0.5, 0.785,0.84) (0.36, 0.4, 0.755, 0.825) (0.36, 0.4, 0.755, 0.825) 0.406 0.647* 0.585 0.585
M2 (0.36, 0.4, 0.645, 0.695) (0.465, 0.5, 0.675, 0.71) (0.36, 0.4, 0.645, 0.695) (0.36, 0.4, 0.645, 0.695) 0.525 0.587* 0.525 0.525
M3 (0.23, 0.305,0.68, 0.74) (0.36, 0.415, 0.79, 0.845) (0.155, 0.21, 0.57, 0.625) (0.36, 0.415, 0.79, 0.845) 0.489 0.602* 0.390 0.602*
Macro environment (0.3449, 0.4077, 0.7782, 0.8387) (0.325, 0.3943, 0.7741, 0.8366) (0.34, 0.4048, 0.748, 0.8222) (0.3619, 0.42, 0.7813, 0.8416) 0.592 0.583 0.584 0.601*
ME1 (0.305, 0.355, 0.695, 0.77) (0.175, 0.245, 0.585, 0.665) (0.375, 0.43, 0.755, 0.825) (0.375, 0.43, 0.755, 0.825) 0.531 0.417 0.596* 0.596*
ME2 (0.36, 0.415, 0.79, 0.845) (0.36, 0.415, 0.79, 0.845) (0.23, 0.305, 0.68, 0.74) (0.36, 0.415, 0.79, 0.845) 0.602* 0.602* 0.489 0.602*
ME3 (0.335, 0.415, 0.74, 0.805) (0.335, 0.415, 0.74, 0.805) (0.26, 0.32, 0.63, 0.69) (0.26, 0.32, 0.63, 0.69) 0.574* 0.574* 0.475 0.475

* denotes the best alternative for the considered main attribute or sub-attribute.

3951
3952 T. Demirel et al. / Expert Systems with Applications 37 (2010) 3943–3952

Z Auephanwiriyahul, S., Keller, J. M., & Gader, P. D. (2002). Generalized Choquet fuzzy
þ
ðCÞ faþ¼0 dg a¼0 ¼ 0:8448: integral fusion. Information Fusion, 3, 69–85.
Cheng, E. W. L., & Li, H. (2004). Exploring quantitative methods for project location
selection. Building and Environment, 39, 1467–1476.
That is, Colson, G., & Dorigo, F. (2004). A public warehouses selection support system.
Z European Journal of Operational Research, 153, 332–349.
ðCÞ ~f dg~ ¼ ½0:353; 0:8448: Current, J., Min, H., & Schilling, D. (1990). Multiobjective analysis of facility location
decisions. European Journal of Operational Research, 49, 295–307.
Delgado, M., Herrera, F., Herrera-Viedma, E., & Martnez, L. (1998). Combining
In Table 9, using the calculation for Choquet integral just above, the numerical and linguistic information in group decision making. Information
overall location values are obtained. The defuzzified overall values Sciences, 107, 177–194.
of alternative locations using generalized Chouqet Integral are also Drezner, Z. (1995). Facility location: A survey of applications and methods. New York:
Springer.
given in the same table . For location 1, the overall Choquet integral Drezner, Z., Scott, C., & Song, J. S. (2003). The central warehouse location problem
value at a ¼ 0 are found as follows: revisited. IMA Journal of Management Mathematics, 14, 321–336.
Geoffrion, A. M., & Graves, G. W. (1974). Multicommodity distribution system
k ¼ 0:99; gðAð5Þ Þ ¼ 0:72; gðAð4Þ Þ ¼ 0:926; gðAð3Þ Þ ¼ 0:97; design by Benders decomposition. Management Science, 2, 82–114.
Grabisch, M., Murofushi, T., & Sugeno, M. (2000). Fuzzy measures and integrals. New
gðAð2Þ Þ ¼ 1:0 and gðAð1Þ Þ ¼ 1:0 York: Physica-Verlag.
R Hidaka, K. K., & Okano, H. (1997). Simulation-based approach to the warehouse
and finally. ðCÞ ~f dg~ ¼ ½0:449; 0:844 location problem for a large-scale real instance. In Proceedings of the 1997 winter
From Table 9, the defuzzified overall values of alternative loca- simulation conference.
tions using generalized Chouqet Integral are obtained as 0.640, Ishii, K., & Sugeno, M. (1985). A model of human evaluation process using fuzzy
integral. International Journal of Man–Machine Studies, 22(1), 19–38.
0.653, 0.605, and 0.625. This means that the ranking order from Kahraman, C., Ertay, T., & Büyüközkan, G. (2006). A fuzzy optimization model for
the best to the worst is Izmit/Gebze, Izmir, Bursa, and Zonguldak. QFD planning process using analytic network approach. European Journal of
The best alternative Izmit/Gebze has the largest weights for labor Operational Research, 171, 390–411.
MacCarthy, B. L., & Atthrirawong, W. (2003). Factors effecting location decisions in
characteristics, infrastructure, and markets, while Izmir is for infra-
international operations – A Delphi study. International Journal of Operations and
structure; Bursa is for costs and macro environments, and Zonguldak Production Management, 23(7), 794–818.
is for none. Izmir and Izmit/Gebze have the same importance in Marichal, J.-R., Meyer, P., & Roubens, M. (2005). Sorting multi-attribute alternatives:
The TOMASO method. Computers and Operations Research, 32, 861–877.
terms of costs and infrastructure. Similar comments can be made
Meyer, P., & Roubens, M. (2006). On the use of the Choquet integral with fuzzy
analyzing the results in Table 9. numbers in multiple criteria decision support. Fuzzy Sets and Systems, 157,
927–938.
Michel, L., & Hentenryck, P. V. (2004). A simple tabu search for warehouse location.
6. Conclusion European Journal of Operational Research, 157, 576–591.
Min, H., & Melachrinoudis, E. (1999). The relocation of a hybrid manufacturing/
The Choquet integral is an extension of the additive operator for distribution facility from supply chain perspectives: A case study, omega. The
International Journal of Management Science, 27, 75–85.
the multi-criteria decision-making problems, but it considers Modave, F., & Grabisch, M. (1998). Preference representation by the Choquet
interactions between criteria. Taking into consideration informa- integral: The commensurability hypothesis. In Proceedings 7th international
tion fusion between criteria and linguistic terms, this study uses conference on information processing and management of uncertainty in
knowledge-based systems (IPMU), Paris, France, July 1998.
the generalized Choquet integral. In the calculation of the Choquet
Owen, S. H., & Daskin, M. S. (1998). Strategic facility location: A review. European
integral, fuzzy measures play an important role, although their Journal of Operational Research, 111, 423–447.
computation processes are quite complicated. A warehouse loca- Partovi, F. Y. (2006). An analytic model for locating facilities strategically, omega.
The International Journal of Management Science, 34, 41–55.
tion selection is a multi-criteria decision-making problem includ-
Saaty, T. L. (1996). Decision makings with dependence and feedback: The analytic
ing both quantitative and qualitative main and sub-criteria. This network process. Pittsburgh, PA: RWS Publications.
paper shows that when these criteria include interactions between Sharma, R. R. K. (1991). Modeling a fertilizer distribution system. European Journal
each others, Choquet integral presents an excellent tool for the of Operational Research, 51, 24–34.
Sharma, R. R. K., & Berry, V. (2007). Developing new formulations and relaxations of
solution. Since a hierarchical multi-criteria problem can be han- single stage capacitated warehouse location problem (SSCWLP): Empirical
dled by generalized Chouqet integral, it is an alternative method investigation for assessing relative strengths and computational effort.
against fuzzy ANP, which can also handle the dependent criteria European Journal of Operational Research, 177(2), 803–812.
Sharma, R. R. K., & Sharma, K. D. (2000). A new dual based procedure for the
and hierarchical problem structure. The warehouse location selec- transportation problem. European Journal of Operational Research, 122(3),
tion problem in this paper can be solved by fuzzy ANP and the ob- 96–109.
tained results can be compared for further research. Shee, D. Y., Tzeng, G. H., & T, I. (2003). AHP fuzzy measure and fuzzy integral
approaches for the appraisal of information service providers in Taiwan. Journal
of Global Information Technology Management, 6(1), 8–30.
References Sugeno, M. (1974). Theory of fuzzy integrals and its applications. Ph.D. thesis, Tokyo
Institute of Technology, Tokyo.
Alberto, P. (2000). The logistics of industrial location decisions: An application of Tsai, H.-H., & Lu, I.-Y. (2006). The evaluation of service quality using generalized
the analytical hierarchy process methodology. International Journal of Logistics: Choquet integral. Information Sciences, 176(6), 640–663.
Research and Application, 3(3), 273–289. Vlachopoulou, M., Silleos, G., & Manthou, V. (2001). Geographic information
Arostegui, M. A., Kadipasaoglu, S. N., & Khumawala, B. M. (2006). An empirical systems in warehouse site selection decisions. International Journal of
comparison of Tabu search, simulated annealing, and genetic algorithms for Production Economics, 71, 205–212.
facilities location problems. International Journal of Production Economics,
103(2), 742–754.

You might also like