Expert Systems with Applications 38 (2011) 3346–3352

Contents lists available at ScienceDirect

Expert Systems with Applications

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

A fuzzy AHP-DEA approach for multiple criteria ABC inventory classification

A. Hadi-Vencheh a,⇑, A. Mohamadghasemi b
a
Department of Mathematics, Islamic Azad University, Khorasgan Branch, Isfahan, Iran
b
Department of Management, Islamic Azad University, Zabol Branch, Iran

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

Keywords: In order to efficiently control the inventory items and determine the suitable ordering policies for them,
Multiple criteria ABC inventory multi-criteria ABC inventory classification, which is one of the most common techniques of production
classification and inventory control, is used. In this classification, other criteria in addition to annual dollar usage
Fuzzy analytic hierarchy process are taken into account and then the items are classified in three classes with different ordering policies,
Data envelopment analysis
based on their priority. In this paper, we propose an integrated fuzzy analytic hierarchy process-data
Simple additive weighting
envelopment analysis (FAHP-DEA) for multiple criteria ABC inventory classification. The proposed
FAHP–DEA methodology uses the FAHP to determine the weights of criteria, linguistic terms such as Very
High, High, Medium, Low and Very Low to assess each item under each criterion, the data envelopment
analysis (DEA) method to determine the values of the linguistic terms, and the simple additive weighting
(SAW) method to aggregate item scores under different criteria into an overall score for each item. The
integrated FAHP–DEA methodology is illustrated using a real case study.
 2010 Elsevier Ltd. All rights reserved.

1. Introduction criticality of part, lead time, commonality, obsolescence, reparabil-

ity, number of requests, scarcity, durability, perish ability, repara-
In an organization, even with small size, hundreds of items may bility, demand distribution, stock ability are also needed for
be held in a warehouse. Controlling all these items by means of classification (Chen, Li, Kilgour, & Hipel, 2008; Cohen & Ernst,
tight ordering policies is not rational in terms of both economy 1988; Flores & Whybark, 1986, 1987; Ng, 2007; Partovi &
and time limitation and will be resulted in complexity the work Anandarajan, 2002; Ramanathan, 2006; Zhou & Fan, 2007; Hadi-
of managers and extra costs. In addition, adopting the inexact Vencheh, 2010). Depending on the nature of items and industry,
ordering policies for items with high priority may be faced the these criteria have different weights. In the real world, prioritizing
stock-out of inventory, thereby losing the market share. On the of the weights of criteria is always as subjective. In other words,
other hands, the tight ordering policies for items with low priority depending on the conditions governing on industry and market,
will generate high inspection costs, and as a consequence, the addi- inventory managers assign different weights to the criteria. For
tional costs will be imposed into the inventory system of organiza- example, when the suppliers ensure that they will provide the re-
tion. Thus, an efficient ordering policy, in view of the priority of quired items in due time, the weight of criterion of lead time is
item, is always making attempts to solve these two situations in lower than other criteria in their opinion. Also, the importance of
such a way that minimizes the holding and inspection costs of criteria differs for different industries. For a perishable item relat-
inventory on the one hand, and prevents stock-out caused by lack ing to food industry, expiration date is an important criterion,
of inventory on the other hand. To achieve such a success, the clas- whereas it may not be very important for the vehicle parts manu-
sifying items inside the groups with different priority should first factures. Thus, we need a model that, on the one hand, meets these
be acted and then, suitable ordering policies should be adopted needs and, on the other hand, any number of qualitative criteria
for each group. ABC classification is one of the most common tech- can be added to that for classification.
niques of classification, dividing items into 3 classes, namely, A A great article has been written on Multiple Criteria ABC (MC-
(very important), B (moderately) and C (least important), based ABC) classification with their points of strenghts and weakness.
on Pareto principle. Traditional ABC (TABC) only uses the criterion Flores and Whybark (1986, 1987) proposed the bi-criteria matrix
of annual dollar usage, but many papers have mentioned that in approach, wherein annual dollar usage by a joint-criteria matrix
addition to this criterion, such other criteria as ordering cost, is combined with another criterion. Though this approach is inter-
esting, it accompanies some limitations. Their approach can not
use three or more criteria to classify inventory items and also
⇑ Corresponding author. weights of all criteria taken into account equal. Chen et al. (2008)
E-mail address: [email protected] (A. Hadi-Vencheh). proposed a case-based distance model for multiple criteria ABC

0957-4174/$ - see front matter  2010 Elsevier Ltd. All rights reserved.
doi:10.1016/j.eswa.2010.08.119
 A. Hadi-Vencheh, A. Mohamadghasemi / Expert Systems with Applications 38 (2011) 3346–3352 3347

analysis, which their approach has been arisen from Flores et al. cation of the proposed FAHP–DEA methodology to a real case
method (Flores & Whybark, 1986, 1987). Advantage of this model study. Section 5 concludes.
is that is easily considered any finite number of criteria for classi-
fication. In this model, criteria weights and sorting thresholds are 2. Fuzzy sets theory
generated mathematically based on the decision maker’s assess-
ment of a set the cases. But information cases are very important Theory of fuzzy sets is quite similar to man’s attitude when fac-
and if this information is incorrect affect process of classification ing uncertainties to express inaccurate words, such as ‘‘approxi-
other items, also its learning may be difficult for the average man- mately”, ‘‘very”, ‘‘nearly”, etc. as well as for consistency with
ager. Partovi and Anandarajan (2002) proposed an artificial neural subjective judgments of different people due to various interpreta-
network (ANN) approach for inventory classification. In their ap- tions from a subject. Zadeh (1965) introduced fuzzy sets theory for
proach two type of learning method, namely back propagation the first time, expressing it in the issue of decision-making. In fuz-
and generic algorithms are used to examine the ANN classification zy sets, membership degree of an element is between 0 and 1,
power and then their results are compared with together. Their ap- while in classic sets, there are two states: an element with the de-
proach finds and bringing out nonlinear relationships and interac- gree 1 is inside the set, or it is not with degree 0. In order to elab-
tions between criteria. However, as authors have asserted, number orate on the said matter, consider the discussion in this paper, in
of criteria are restricted, also entering many qualitative criteria which MC-ABC inventory classification is carried out using inven-
into model may be difficult and in addition, learning their tory managers’ subjective judgments and introducing fuzzy con-
meta-heuristics approach is difficult for inventory managers. cepts of prioritizing the criteria. To achieve these ends, fuzzy set,
Ramanathan (2006) proposed a weighted linear optimization fuzzy numbers and linguistic variables should first be introduced
model for multiple criteria ABC inventory classification, where per- (Chen, 2000; Kaufmann & Gupta, 1991; Zadeh, 1965;
formance score of each item obtained using a DEA-like model. Zimmermann, 1987, 1991).
However his model may result in a position in which an item with
a high value in an unimportant criterion is inappropriately classi- Definition 2.1. A fuzzy set Ae in a universe of discourse X is defined
fied as class A. This drawback was rectified by Zhou and Fan by a membership function le ðxÞ which associates "x 2 X a real
A
(2007) via obtaining most favorable and least favorable scores for number in the interval [0,1]. le ðxÞ express membership degree of x
A
each item. Ng (2007) proposes a weighted linear model for MC- e
in A.
ABC inventory classification. Via a proper transformation, the Ng-
model can obtain the scores of inventory items without a linear
optimizer. The Ng-model is simple and easy to understand. Despite e is a crisp set
Definition 2.2. The a-cut of fuzzy set A
its many advantages, Ng-model leads to a situation which the e e
A a ¼ fxjle ðxÞ P ag. The support A is the crisp set
A
weight of an item may be ignored. To overcome this drawback e ¼ fxjl ðxÞ P 0g. A
Suppð AÞ e is normal if and only if
eA
Hadi-Vencheh (2010) proposed a simple nonlinear programming
Suppx2X leðxÞ ¼ 1.
model which determines a common set of weights for all the items. A
Inventory classification is essentially a multiple criteria decision
making (MCDM) problem, which involve multiple assessment cri- Definition 2.3. A fuzzy subset Ae of universe set X is convex if and
teria such as annual dollar usage, average unit cost, lead time and only if le ðkx þ ð1  kÞyÞ P minðle ðxÞ; le ðyÞÞ; 8x; y 2 X; k 2 ½0; 1,
A A A
so on. Therefore, MCDM approaches can be used for inventory clas- where min denotes the minimum operator.
sification. Of the MCDM approaches, the analytic hierarchy process
(AHP) method is particularly suitable for modeling qualitative cri-
e is a fuzzy number if and only if A
Definition 2.4. A e is normal and
teria and has found extensive applications in a wide variety of
areas such as selection, evaluation, planning and development, convex fuzzy set of X.
decision making forecasting, and so on. However, due to the fact
that there are tenths or hundreds of items to be evaluated and pri- Definition 2.5. A triangular fuzzy number Ae is defined with piece-
oritized, while the AHP method can only compare a very limited wise linear membership function le ðxÞ as follow:
A
number of decision alternatives, the pair-wise comparison manner 8 xa1
is obviously infeasible in this situation. To overcome this difficulty, > ; a1 6 x 6 a2 ;
< a2 a1
we combine the AHP with the data envelopment analysis (DEA) leA ðxÞ ¼ aa33a
x
; a2 6 x 6 a3 ;
>
: 2
and propose an integrated AHP–DEA methodology in this paper. 0; otherwise;
On the other hand, fuzzy variables are highly suitable for express-
ing of decision makers’ subjective judgments on the issues which And as a triplet (a1, a2, a3) is indicated, where a1, a3 the lower and
have both qualitative and quantitative variables. These variables upper bounds respectively, and a2 is the most likely value of A. e
apply fuzzy numbers for prioritizing and ranking. In this paper,
these variables are used to prioritize criteria, and ranking the Definition 2.6. A positive trapezoidal fuzzy number C e can be
items, in view of the measures of a portion of items with respect defined as (a1, a2, a3, a4) and its membership function is defined as
to the criterion, a suitable linguistic variable is selected. For exam- 8
ple, for classification of items, if the measure of an item is high for > 0; x 6 a1 ;
>
>
the annual dollar usage (refer to Section 4.4.2), it is said that the > xa1 ;
> a1 6 x 6 a2 ;
>
< a2 a1
measure of this item is high. As we will see, in the discussion of leC ðxÞ ¼ 1; a2 6 x 6 a3 ;
MC-ABC classification, using the fuzzy variable, especially to assert >
>
>
> a4 x
; a3 6 x 6 a4 ;
the measure of the qualitative criteria, will contribute to the inven- >
> a a
: 4 3
tory mangers’ decision-making. 0; a4 P x;
This paper is organized as follows. In Section 2, we give a brief
description of the fuzzy set theory and provide a ground for the la- Definition 2.7. Let Ae ¼ ða1 ; a2 ; a3 Þ and B
e ¼ ðb1 ; b2 ; b3 Þ be two posi-
ter development of methodology. In Section 3, we develop an inte- tive triangular fuzzy numbers and r be a positive real number.
grated FAHP–DEA methodology for the MCDM problems with a Then, sum, multiplication, subtraction, distance and inversion of
large number of decision alternatives. Section 4 presents an appli- these two triangular fuzzy number is defined as follow
3348 A. Hadi-Vencheh, A. Mohamadghasemi / Expert Systems with Applications 38 (2011) 3346–3352

eB
A e ¼ ½a1 þ b1 ; a2 þ b2 ; a3 þ b3 ; maximize l;
e
AB e ¼ ½a1  b1 ; a2  b2 ; a3  b3 ; subject to ðmru  lru Þlwu  wr þ lru wu 6 0;
eB e ¼ ½a1  b3 ; a2  b2 ; a3  b1 ; ðuru  mru Þlwu þ wr  uru wu 6 0;
A ð1Þ
rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi X
C
h i wj ¼ 1; wj > 0; j ¼ 1; . . . C;
e BÞ
dð A; e ¼ 1 ða1  b1 Þ2 þ ða2  b2 Þ2 þ ða3  b3 Þ2 ;
3 j¼1

e  r ¼ ½a1  r; a2  r; a3  r;
A r ¼ 1; . . . ; C  1; u ¼ 2; . . . ; C; u > r;
 
e 1  ¼ 1 ; 1 ; 1 :
½A where C is the number of criteria, lru, uru and mru are the lower,
a3 a2 a1
upper bounds and most likely value of triangular fuzzy numbers
in pairwise comparisons matrix, respectively, when experts com-
3. The proposed model pare rth criterion (alternative) with respect to uth criterion (alterna-
tive) and their values have been presented in Table 1, wj is the
3.1. Determination of the weights of criteria using FAHP weight of criterion j and l is the consistency index. If l = 1 then
the fuzzy pairwise comparison matrix is said to be consistent, and
One of the important issues of decision-making is prioritization if l is negative then the matrix is said to be strongly inconsistent.
of criteria. Determining the importance of weights by managers, On the other hand, if l is close to 1 then the fuzzy pairwise compar-
especially in terms of issue of MC-ABC classification, is always sub- ison matrix is said to be almost consistent. The final purpose it to
jective in such a way that inventory managers usually select some aggregate multiple performance scores of an item with respect to
important criteria and then prioritize them. There are several different criteria into a single score for the subsequent ABC inven-
methods to determine of the criteria weights, including analytic tory classification. The score of each item is defined using the fol-
hierarchy process (or fuzzy AHP), entropy analysis, eigenvector lowing simple additive weighting (SAW) method (Hwang & Yoon,
method, weighted least square method and linear programming 1981):
for multidimensions of analysis preference (LINMAP). In this paper,
X
C
we apply the method of fuzzy (FAHP). The AHP was first intro- Sr ¼ wrj wj ; r ¼ 1; . . . ; R; ð2Þ
duced by Saaty (1980), where a MCDM problem is transformed j¼1
into a hierarchical structure. Fig. 1 show a hierarchical analysis
structure for MC-ABC inventory classification. where R is the number of items, wj is the weight of criterion j ob-
Generally, it is impossible that crisp values reflect the decision tained by solving nonlinear optimization problem and wrj is the
maker’ uncertain preferences. Therefore, FAHP is proposed to re- weight of item r with respect to criterion j which is dealt with in de-
lieve the uncertainness of AHP method, where the fuzzy compar- tail in the next section and Sr is the total score of item r.
isons ratios are used. Also, there are the several procedures to
attain the priorities in FAHP. The fuzzy least square method (Xu,
2000), method based on the fuzzy modification of the LLSM 3.2. The FAHP-DEA methodology
(Boender, de Graan, & Lootsma, 1989), geometric mean method
(Buckley, 1985), the direct fuzzification of the method of Csutora Consider a generic MC-ABC inventory classification problem
and Buckley (2001), synthetic extend analysis (Chang, 1996), with C criteria and R items, where any decision criterion can be fur-
Mikhailov’s fuzzy preference programming (Mikhailov, 2003) ther broken down into more sub-criteria if necessary. The normal-
and two-stage logarithmic programming (Wang, Yang, & Xu, ized weight vector of decision criteria, W = (w1, . . . , wC), is assumed
2005) are some of these methods. We apply Mikhailov’s fuzzy to have been obtained through Mikhailov’s method (Mikhailov,
preference programming (Mikhailov, 2003), since, it is able to de- 2003). To characterize the relative importance of each item with
rive inconsistency value and crisp priorities vector of pairwise respect to each criterion, we define for each criterion a set of
comparison matrices concurrently. Mikhailov (2003) use both assessment grades Gj = {Pj1, . . . , PjN} (j = 1, . . . , C), where Pj1, . . . , PjN
linear and nonlinear optimization methods, which maximize the represent the importance from the most to the least important
decision maker’s satisfaction on the derived weight vector. In and N is the number of assessment grades for criterion j. We then
the proposed model first we obtain the prioritization vector by ask the experts to assess the items and classify them into their cor-
solving the following nonlinear optimization model given by Mik- responding assessment grades in terms of their relative impor-
hailov (2003): tance with respect to the criterion under consideration. Also,
Fig. 2 shows a typical Membership Function (MF) constructed by
experts for the measures of items with respect to each criterion
that its application would make easier the selection of grades.
Assume that criterion j will be assessed by X experts (j = 1, . . . , C)
and xrjn are the number of experts who assess item r to grade Pjn
PN
under the criterion j. It is evident that n¼1 xrjn ¼ X.With these
notations, let MC-ABC inventory classification matrix be as Table 2.

Table 1
Fuzzy linguistic variables and their fuzzy numbers, membership function.

Linguistic variables Fuzzy number Membership function

Equally important ~
1 (1, 1, 2)
Weakly important ~
2 (1, 2, 3)
More important ~
3 (2, 3, 4)
Strongly important ~
4 (3, 4, 5)
Absolutely important ~
5 (4, 5, 5)
Fig. 1. A hierarchical structure of AHP for MC-ABC inventory classification.
 A. Hadi-Vencheh, A. Mohamadghasemi / Expert Systems with Applications 38 (2011) 3346–3352 3349

Table 4
The weights of criteria.

Criteria Weight
Annual dollar usage 0.3351
limitation of warehouse space 0.3792
Average lot cost 0.1840
Lead time 0.1017
l 0.83

impose the strong ordering condition w(Pj1) P 2w(Pj2) P

Fig. 2. MF of the measures of items with respect to each criterion.
P Nw(PjN) P 0 on their weights, which ensures that
w(Pj1) P w(Pj2) P P w(PjN) P 0 and w(Pj1)  w(Pj2) P w(Pj2) 
w(Pj3) P P w(PjN1)  w(PjN) P 0.
Table 2 By solving model (4) for each criterion, respectively, the local
MC-ABC inventory classification matrix.
weights of each item with respect to the C decision criteria can
Item Criteria all be generated by Eq. (3) and can then be aggregated using the
C1 Cj CC SAW method into a total overall weight or score that item, as
shown below:
P11 P1N Pj1 PjN PC1 PCN
!
A1 x111 x11N x1j1 x1jN x1C1 x1CN X
C X
C X
N
..
.
..
.
..
.
..
.
..
.
..
.
..
.
Sr ¼ wj wrj ¼ wj 
w ðPjn Þxrjn ; r ¼ 1; . . . ; R; ð5Þ
j¼1 j¼1 n¼1
Ar xr11 xr1N xrj1 xrjN xrC1 xrCN
.. .. .. .. .. .. ..
. . . . . . . where wj(j = 1, . . . , C) are the weights of criteria by the Mikhailov’s
AR xR11 xR1N xRj1 xRjN xRC1 xRCN
fuzzy preference programming (Mikhailov, 2003), w*(Pjn)(j = 1, . . . ,
C; n = 1, . . . , N) are the optimal weights of assessment grades

This table shows the number of experts who select their desired
grade for each item with respect to every criterion.
Let w(Pjn) be the weight of grade Pjn(n = 1, . . . , N). Then, the local
weight of each item with respect to every criterion can be defined
as
X
N
wrj ¼ wðPjn Þxrjn r ¼ 1; . . . ; R; j ¼ 1; . . . ; C: ð3Þ
n¼1

To determine the local weight of each item i with respect to every

criterion, we view each item as a decision making unit (DMU),
w(Pjn) as a decision variable and also the weight assigned to the out-
put xijn, and construct the following DEA model: Fig. 3. MF of the measures of items with respect to annual dollar usage (Thousand
$).
X
N
wij ¼ max wðPjn Þxijn ;
n¼1
X
N ð4Þ
s:t: wðPjn Þxrjn 6 1; r ¼ 1; . . . ; R; j ¼ 1; . . . ; C;
n¼1

wðPj1 Þ P 2wðPj2 Þ P P NwðP jN Þ P 0;

where w(Pj1), . . . , w(PjN) are the decision variable and

w(Pj1) P 2w(Pj2) P P Nw(PjN) P 0 is the strong ordering condi-
tion imposed on grades, which is similar to the strong ordering con-
dition on different ranking places in voting systems proposed by
Noguchi, Ogawa, and Ishii (2002) and Wang, Liu, and Elhag
(2008). Although there are no ranking places and votes in the distri-
bution decision matrix in Table 2, the assessment grades Pj1, . . . , PjN,
which represent the importance from the most to the least
important, are in fact ranking ordered. Thus, it makes sense to Fig. 4. MF of the measures of items with respect to limitation of warehouse space.

Table 3
Pairwise comparisons matrix of criteria.

Criteria Annual dollar usage limitation of warehouse space Average lot cost Lead time
Annual dollar usage (1, 1, 1) (1/2, 1, 1) (1, 2, 3) (2, 3, 4)
limitation of warehouse space (1, 1, 1) (1, 2, 3) (3, 4, 5)
Average lot cost (1, 1, 1) (1, 2, 3)
Lead time (1, 1, 1)
3350 A. Hadi-Vencheh, A. Mohamadghasemi / Expert Systems with Applications 38 (2011) 3346–3352

determined by model (4) and Sr(r = 1, . . . , R) are the scores of the R

inventory items, based upon which the R items can be classified.

4. Case study

4.1. Area of case study and method of data collection

To accredit our model, it is implemented in a factory. This fac-

tory comprises two separate sections. In the first section, there is
soft-drink production line, and in the second one, biscuit produc-
tion line. We implemented our model in the soft drink production
line’s raw material warehouse. This warehouse consists of 35
Fig. 5. MF of the measures of items with respect to average lot cost ($). items. To collect data, we used the point of views of 10 experts.

4.2. Determination of criteria

The experts participating in the implementation of this model

have regarded 4 important criteria for classification of inventory.
These criteria include the following:

C1: Annual dollar usage: This criterion is usually important in

all multiple criteria ABC classification problems.
C2: Lead time: This criterion comprises the time required from
the time of order placement to time of reaching it to the
warehouse.
C3: Average lot cost: Any available item in the warehouse
always faces to costs of ordering and holding, but different
Fig. 6. MF of the measures of items with respect to lead time (day). items have different costs. For example, by reason of holding

Table 5
Multiple criteria ABC classification decision making matrix.

Number of item Criteria

Annual dollar usage Lead time Average lot cost Limitation of warehouse space
VH H M VL L VH H M VL L VH H M VL L VH H M VL L
1 0 0 0 1 9 0 0 7 2 1 0 0 2 4 4 0 0 0 0 10
2 0 0 2 4 4 8 2 0 0 0 1 3 5 1 0 0 0 0 0 10
3 0 1 8 1 0 2 3 5 0 0 10 0 0 0 0 0 2 6 2 0
4 0 0 8 2 0 1 3 6 0 0 10 0 0 0 0 0 1 5 4 0
5 10 0 0 0 0 1 5 4 0 0 2 6 2 0 0 7 2 1 0 0
6 10 0 0 0 0 0 1 8 1 0 0 0 5 4 1 0 0 3 3 4
7 6 3 1 0 0 0 1 7 2 0 0 0 1 5 4 0 0 2 4 4
8 9 1 0 0 0 0 2 8 0 0 0 0 1 5 4 0 2 5 3 0
9 0 0 2 5 3 0 1 8 1 0 0 0 2 5 3 0 0 0 3 7
10 0 0 1 3 6 10 0 0 0 0 10 0 0 0 0 7 3 0 0 0
11 0 0 0 3 7 10 0 0 0 0 10 0 0 0 0 8 1 1 0 0
12 0 0 0 0 10 3 2 5 0 0 10 0 0 0 0 0 1 2 2 5
13 0 0 0 0 10 2 3 5 0 0 10 0 0 0 0 0 0 2 4 4
14 0 0 0 0 10 0 0 5 3 2 0 0 0 2 8 0 0 0 0 10
15 0 0 0 0 10 0 0 6 2 2 0 0 0 0 10 0 0 0 0 10
16 0 1 9 0 0 0 0 7 2 1 0 0 0 0 10 0 0 1 2 7
17 10 0 0 0 0 0 0 0 3 7 1 2 7 0 0 10 0 0 0 0
18 0 0 5 4 1 7 2 1 0 0 0 2 7 1 0 0 0 1 2 7
19 0 0 5 4 1 0 0 1 2 7 0 0 5 4 1 0 0 1 3 6
20 5 3 2 0 0 2 4 4 0 0 0 1 7 2 0 0 0 2 3 5
21 0 0 0 1 9 0 0 1 1 8 0 0 2 6 2 0 0 0 0 10
22 0 0 6 2 2 0 0 1 3 6 0 0 3 5 2 0 0 0 4 6
23 0 1 7 2 0 10 0 0 0 0 0 1 6 3 0 0 0 0 2 8
24 0 0 6 3 1 10 0 0 0 0 0 0 2 5 3 0 0 0 0 10
25 0 0 3 5 2 9 1 0 0 0 0 0 2 4 4 0 0 0 1 9
26 0 0 0 0 10 10 0 0 0 0 0 0 1 4 5 0 0 0 1 9
27 0 0 0 0 10 10 0 0 0 0 0 0 2 2 6 0 0 0 0 10
28 0 0 0 3 7 10 0 0 0 0 0 0 3 5 2 0 0 0 0 10
29 0 0 0 0 10 10 0 0 0 0 0 0 1 6 3 0 0 0 1 9
30 0 0 3 4 3 10 0 0 0 0 0 0 4 4 2 0 0 0 0 10
31 0 0 0 0 10 10 0 0 0 0 0 0 1 6 3 0 0 0 0 10
32 0 0 1 7 2 0 0 8 2 0 0 0 0 1 9 0 0 0 2 8
33 0 0 0 2 8 10 0 0 0 0 0 0 1 1 8 0 0 0 0 10
34 0 0 0 1 9 3 3 4 0 0 10 0 0 0 0 0 0 1 2 7
35 0 0 2 6 2 0 0 6 3 1 2 5 3 0 0 0 0 2 6 2
 A. Hadi-Vencheh, A. Mohamadghasemi / Expert Systems with Applications 38 (2011) 3346–3352 3351

in special condition (holding in the freezing conditions), some consistency index is close to 1, experts decided that these weights
of those have the holding cost and should be noted. would be employed in the next part.
C4: Limitations of warehouse space: Due to available limita-
tions of stocking of items in the warehouse, this criterion 4.4. Classification
should also be taken into account. For instance, due to the spe-
cial characteristic of an item, they cannot be put on each other 4.4.1. Fuzzification of criteria
infinitely, since heaviness of weight causes damage to them To simplify specialists’ appraisal in selecting grades, the mea-
due to putting a large number of them on each other, and this sures of items with respect to criteria should be necessarily fuzz-
situation consequently causes to increase space in the ware- ified. To this end, measures of these criteria are changed into
house. Thus, this group of items should be investigated more linguistic variables, defined as high, medium and low. This task
accurately. will has been led to become more simplicity the experts’ appraisal
with respect to of greatness of measures of different criteria. Trap-
As show, the above four criteria are positively related to score of ezoidal membership functions of these linguistic variables are
an item. It means that the larger measure an item with respect to determined based on experts’ point of views over data in records
criterion j has higher priority as compared to other items. Whereas of items. (For the qualitative criterion, i.e. limitation of warehouse
the fourth criterion is a qualitative criterion and also, by reason of space, these membership functions emerge based on experts’ judg-
governing conditions on market, measures of some criteria, such as ments and interval of amounts for this criterion, are assumed
lead time, may change during a year. Therefore, in order to make based on the experts’ decision arbitrarily from 0 to 100.) These
easier the appraisal process, experts decided to use MF of low, functions are displayed in Figs. 3–6.
medium and high (Section 4.4.1). Measures of criterion of limita-
tion of warehouse space are put at 0 to 100 based on the experts’
4.4.2. Definition of grades and forming a MC-ABC classification
point of views.
decision making matrix
To appraise the measure of items with respect to criteria, we
4.3. Prioritization of criteria need to define a set of all grades for each one of criteria. To this
end, we use the following grades for each criterion.
After determining the most important criteria, experts form
pairwise comparisons of the criteria in Table 3 using the set of fuz- Gj ¼ fVery High; High; Medium; Low; Very Lowg
zy linguistic variables and their membership functions in Table 1 ¼ fVH; H; M; L; VLg; ðj ¼ 1; . . . ; CÞ;
unanimously, and then by using Mikhailov nonlinear optimization
model (Mikhailov, 2003), crisp weight vector of these criteria are If required, the number of grades for different criteria can be chan-
directly achieved. As displayed in Table 4, since the quantitative ged. In this paper, we use any five grades with respect to each

Table 6
The local and overall weights of items and comparing to TABC.

Number of item Local weights Scores Proposed model Traditional

Annual dollar usage Lead time Average lot cost Limitation of warehouse space
17 1.0000 0.2150 0.4333 1.0000 0.8156 A A
5 1.0000 0.4833 0.5667 0.8333 0.8042 A A
11 0.2150 1.0000 1.0000 0.8833 0.6923 A C
10 0.2283 1.0000 1.0000 0.8500 0.6842 A C
8 0.9500 0.3667 0.2383 0.3417 0.5288 A A
6 1.0000 0.3417 0.2867 0.2550 0.5191 A A
3 0.3417 0.5167 1.0000 0.3500 0.4835 A B
4 0.3167 0.4500 1.0000 0.3167 0.4557 B B
20 0.7167 0.5333 0.3333 0.2417 0.4472 B A
7 0.7833 0.3333 0.2383 0.2467 0.4336 B A
12 0.2000 0.5667 1.0000 0.2667 0.4095 B C
13 0.2000 0.5167 1.0000 0.2467 0.3969 B C
34 0.2050 0.5833 1.0000 0.2233 0.3964 B C
23 0.3333 1.0000 0.3250 0.2100 0.3526 B A
18 0.2867 0.9733 0.3583 0.2233 0.3454 B B
2 0.2467 0.9000 0.4417 0.2000 0.3311 B B
24 0.2950 1.0000 0.2517 0.2000 0.3225 B B
30 0.2600 1.0000 0.2733 0.2000 0.3147 B C
35 0.2567 0.2950 0.5500 0.2567 0.3144 C B
25 0.2650 0.9500 0.2467 0.2050 0.3083 C B
28 0.2150 1.0000 0.2650 0.2000 0.2945 C C
29 0.2000 1.0000 0.2433 0.2050 0.2910 C C
26 0.2000 1.0000 0.2333 0.2050 0.2892 C C
31 0.2000 1.0000 0.2433 0.2000 0.2891 C C
27 0.2000 1.0000 0.2367 0.2000 0.2879 C C
33 0.2100 1.0000 0.2183 0.2000 0.2879 C C
16 0.3500 0.3033 0.2000 0.2233 0.2694 C B
19 0.2867 0.2233 0.2867 0.2283 0.2580 C B
22 0.2900 0.2283 0.2650 0.2200 0.2524 C B
9 0.2517 0.3417 0.2517 0.2150 0.2468 C C
32 0.2483 0.3167 0.2050 0.2100 0.2326 C B
1 0.2050 0.3033 0.2467 0.2000 0.2206 C C
21 0.2050 0.2183 0.2567 0.2000 0.2138 C C
14 0.2000 0.2817 0.2100 0.2000 0.2100 C C
15 0.2000 0.2900 0.2000 0.2000 0.2090 C C
3352 A. Hadi-Vencheh, A. Mohamadghasemi / Expert Systems with Applications 38 (2011) 3346–3352

criterion. After defining these grades, as is shown in Table 5, we of their total scores. A real example investigated to illustrate the
make a MC-ABC classification decision making matrix. Raw data applications of the proposed FAHP–DEA methodology. It is shown
of this matrix comprise number of experts who select their desired that the new integrated FAHP–DEA methodology is simple enough,
grade based on whether measure of a special item with respect to easy to use, applicable to any number of decision alternatives, and
each criterion is high, medium or low, using trapezoidal member- particularly useful and effective to complex MCDM problems with
ship functions. For example, since membership function of the mea- a large number of decision alternatives, where pairwise compari-
sure of item 17 with respect to the criterion of annual dollar usage sons are certainly impossible to be made.
is high, every 10 experts selected high grade for it. The number of
experts for selecting grades in all criteria is equal, i.e. X = 10 experts.
References
4.4.3. Determination of weights
Boender, C. G. E., de Graan, J. G., & Lootsma, F. A. (1989). Multi-criteria decision
Using data of Table 5, we obtained local weight of each item r analysis with fuzzy pairwise comparisons. Fuzzy Sets and Systems, 29(2),
with respect to every criterion by model (4). These local weights 133–143.
are presented in columns 2–5 of Table 6. Besides, the score (total Buckley, J. J. (1985). Fuzzy hierarchical analysis. Fuzzy Sets and Systems, 17(3),
233–247.
weight) of each item was obtained using Eq. (5) and shown in Chang, D. Y. (1996). Applications of the extent analysis method on fuzzy AHP.
the sixth column of Table 6. European Journal of Operational Research, 95(3), 649–655.
Chen, C. T. (2000). Extensions of the TOPSIS for group decision-making under fuzzy
environment. Fuzzy Sets and Systems, 114, 1–9.
4.4.4. Selection of classification of each item using the scores
Chen, Y., Li, K. W., Kilgour, D. M., & Hipel, K. W. (2008). A case-based distance model
At this stage, scores of each item indicates the position of that for multiple criteria ABC analysis. Computers and Operations Research, 35,
item among others. Then, experts decided that the number of 776–796.
items in classes A, B and C is put according to the same number Cohen, M. A., & Ernst, R. N. (1988). Multi-item classification and generic inventory
stock control policies. Production and Inventory Management Journal, 29(3), 6–8.
of items in TABC method, i.e. 7 items for class A, 11 items for class Csutora, R., & Buckley, J. J. (2001). Fuzzy hierarchical analysis: The Lambda-Max
B and 17 items for class C. Therefore, items with scores greater method. Fuzzy Sets and Systems, 120, 181–195.
than or equal to 0.48 are classified as class A, items with scores be- Flores, B. E., & Whybark, D. C. (1986). Multiple criteria ABC analysis. International
Journal of Operations and Production Management, 6(3), 38–46.
tween 0.48 and 0.3145 are classified as class B and the other items Flores, B. E., & Whybark, D. C. (1987). Implementing multiple criteria ABC analysis.
are classified as class C. Selected classes are presented in seventh Journal of Operations Management, 7(1–2), 79–85.
column of Table 6. Hadi-Vencheh, A. (2010). An improvement to multiple criteria ABC inventory
classification. European Journal of Operational Research, 201(3), 962–965.
It is clear from the priority vector of the criteria in Table 4, the Hwang, C. L., & Yoon, K. (1981). Multiple attribute decision making: Methods and
criteria of limitation of warehouse space and average lot cost, application. Berlin: Springer.
which are not considered in the process of TABC, have high weights Kaufmann, A., & Gupta, M. M. (1991). Introduction to fuzzy arithmetic: Theory and
applications. New York: Van Nostrand Reinhold.
and the proposed method considers these criteria. Compared with Mikhailov, L. (2003). Deriving priorities from fuzzy pairwise comparison judgments.
the TABC, only 19 items of the suggested model remain in the same Fuzzy Sets and Systems, 134(3), 365–385.
classes. In other words, by implementing the suggested model, 4 Ng, W. L. (2007). A simple classifier for multiple criteria ABC analysis. European
Journal of Operational Research, 177, 344–353.
out of 7 of class A in the TABC classification remain in the same
Noguchi, H., Ogawa, M., & Ishii, H. (2002). The appropriate total ranking method
class, and other three items are re-classified in class B. Out of 11 using DEA for multiple categorized purposes. Journal of Computational and
in class B, 4 remained in the same class B, 1 is transferred to class Applied Mathematics, 146(1), 155–166.
A and 6 to class C. Moreover, out of 17 items of class C in the TABC Partovi, F. Y., & Anandarajan, M. (2002). Classifying inventory using an artificial
neural network approach. Computers and Industrial Engineering, 41, 389–404.
classification, 11 remained in the same class C, 2 were re-classified Ramanathan, R. (2006). ABC inventory classification with multiple-criteria using
to class A and 4 to class B. weight linear optimization. Computers and Operations Research, 33(3), 695–700.
Saaty, T. L. (1980). The analytic hierarchy process: Planning, priority setting. New York:
McGraw Hill International Book Co.
5. Conclusion Wang, Y. M., Liu, J., & Elhag, T. M. S. (2008). An integrated AHP–DEA methodology
for bridge risk assessment. Computers & Industrial Engineering, 54, 513–525.
Wang, Y. M., Yang, J. B., & Xu, D. L. (2005). A two-stage logarithmic goal
In this paper, we have proposed a new integrated FAHP–DEA
programming method for generating weights from interval comparison
methodology to classify tenth or hundreds of the inventory items. matrices. Fuzzy Sets Systems, 152, 475–498.
To avoid making a large number of pairwise comparisons, the pro- Xu, R. (2000). Fuzzy least square priority method in the analytic hierarchy process.
Fuzzy Sets and Systems, 112(3), 395–404.
posed FAHP–DEA methodology uses only the Mikhilov model
Zadeh, L. A. (1965). Fuzzy sets. Inform and Control, 8, 338–353.
(Mikhailov, 2003) to determine the weights of criteria, linguistic Zhou, P., & Fan, L. (2007). A note on multi-criteria ABC inventory classification using
terms such as Very High, High and so on to assess items under each weighted linear optimization. European Journal of Operational Research, 182,
criterion, DEA model with common weights to determine the val- 1488–1491.
Zimmermann, H. j. (1987). Fuzzy set, decision making and expert system. Boston:
ues of the linguistic terms, and the simple additive weighting Kluwer Academic Publishers.
method to aggregate the item scores under different criteria into Zimmermann, H. J. (1991). Fuzzy set theory and its applications (2nd ed.). London:
a total score of each item. The items can then be classified in terms Kluwer Academic Publishers.