You are on page 1of 12

A Multi-Level Linguistic Fuzzy Decision Network

Hierarchical Structure Model for Crop Selection

Basem Mohamed Elomda1, Hesham Ahmed Hefny2, Fathy Ashmawy3,


Maryam Hazman1, and Hesham Ahmed Hassan4
1
Central Laboratory for Agriculture Expert Systems (CLAES),
Agriculture Research Center (ARC), Cairo, Egypt
{basem,m.hazman}@claes.sci.eg
2
Institute of Statistical Studies and Research (ISSR), Cairo University, Cairo, Egypt
hehefny@cu.edu.eg
3
Central Laboratory for Design and Statistical Analysis Research (CLDSAR),
Agriculture Research Center (ARC), Cairo, Egypt
f_ashmawy@hotmail.com
4
Faculty of Computers and Information (FCI), Cairo University, Cairo, Egypt
h.hassan@fci-cu.edu.eg

Abstract. Cultivate the best crop from many suitable crops is a complex pro-
cess that faces the decision makers (e.g. farmers, their advisors, and others in
the agricultural sector). Their goal is to select a crop which maximizes the re-
source utilization and in the same time ensures the sustainability for natural
agricultural resources. Selecting such crop for cultivating among many suitable
alternatives crops is a Multiple Criteria Decision Making (MCDM) problem.
Since, the selection for the best decision is dependent in many criteria and hav-
ing dependence and feedback among them. In this paper Linguistic Fuzzy De-
cision Network (LFDN) method is developed and applied to a real case study to
decide the cultivate crop among four crops-namely: Wheat, Corn, Rice, and
Fababean w.r.t given multiple criteria.

Keywords: Multi Criteria Decision Making, Fuzzy Decision Map, Fuzzy Cog-
nitive Map, Linguistic Fuzzy Decision Network, Crop Selection.

1 Introduction

Multiple Criteria Decision Making (MCDM) techniques have been developed during
the mid-1960s. It has been a hot area of research in decision theory, operations re-
search, management science and system engineering [1]. MCDM is the process of
ranking the feasible alternatives and selecting the best one by considering several/
multiple and conflicting criteria [2]. Many techniques or methods have been devel-
oped for solving MCDM problems. In 1996, the Analytic Network Process (ANP) is
developed by Saaty for solving MCDM problems to tackle the dependence and feed-
back problem among criteria in the Analytic Hierarchy Process (AHP) method[3], as
extension of AHP proposed by Satty in 1971 [4].

Springer International Publishing Switzerland 2015 497


D. Filev et al. (eds.), Intelligent Systems2014,
Advances in Intelligent Systems and Computing 323, DOI: 10.1007/978-3-319-11310-4_43
498 B.M. Elomda et al.

In 2006, Fuzzy Decision Maps (FDM) method was introduced [5] to address the
shortcoming of ANP and AHP method. In 2013, Linguistic Fuzzy Decision Network
(LFDN) is proposed to tackle the drawback of FDM method [6, 7]. However, LFDN
method provides only the weight of criteria due to it deals only with two level struc-
tures namely: goal and criteria. But in practical real world situation, MCDM problems
consist of multi-level such as: goal, criteria, sub-criteria, and so on until reach to al-
ternative level. Therefore, the main objective of this paper is to develop LFDN meth-
od to take into account the requirement for multi-level hierarchical structure such as,
objectives, criteria, sub-criteria and alternatives. Thus, Multi-Level LFDN hierar-
chical structure (ML-LFDN) is a model for selecting the best alternative when the
human expert or Decision Maker (DM) has many criteria. Then, we apply the modi-
fied method to the empirical case study to determine the best crop among the four
crops with respect to multiple criteria.
This paper is organized as follows: Section 2 presents Fuzzy Cognitive Map
(FCM). Section 3 presents the Linguistic Fuzzy Decision Network (LFDN) method.
Multi-level LFDN (ML-LFDN) hierarchy structure model is explained in section 4. A
case study is given in section 5. Finally, the conclusion is in section 6.

2 Fuzzy Cognitive Map (FCM)


In 1986, Kosko introduced Fuzzy Cognitive Map (FCM) model [8] as an extension of
cognitive map model proposed by Axelrod in 1976 [9]. FCM supported MCDM
with dependence and feedback [5, 6, 7]. Also, FCM with linguistic values in the form
of Triangular Fuzzy Number (TFN) can be used as a tool to forecast the future state of
the system [10]. A FCM has the topology of fuzzy weighted causal directed graphs
with feedback as in Fig 1. Fig 1 illustrates a graphical representation of a FCM con-
sisting of five concepts (or Criteria) namely C1, C2, C3, C4, and C5. These criteria are
connected by weights. The causal relationships between two concepts (described as
edges or directed arcs) are described using a degree of influence i.e., Wij. Human
experts describe these degrees of influence using fuzzy/linguistic values in the form
of TFN's as in table 1. The influence among criteria is calculated by the fuzzy updat-
ing equation as in Eq. (2) [6, 7]. After drawing linguistic FCM to indicate the influ-
ence among criteria, linguistic FCM as in Fig. 1 is converted to the fuzzy weight/
adjacency matrix as in Eq. (1).

W12 C1 W15
W25
C2 C5
W52
W41
W33 W13
W14 W54
C3 C4
W43

Fig. 1. FCM graph example


A Multi-Level Linguistic Fuzzy Decision Network Hierarchical Structure Model 499

Table 1. Linguistic values and corresponding TFN for causal relations between events [7]

linguistic values TFN


No influence (0,0,0)
Weak influence (W) (0,0,0.25)
Medium Weak influence (MW) (0,0.25,0.5)
Medium influence (M) (0.25,0.5,0.75)
Medium strong influence (MS) (0.5,0.75,1)
Strong influence (S) (0.75,1,1)

(0, 0, 0) W12 W13 W14 W15


(0, 0, 0) (0, 0, 0) (0, 0, 0) (0, 0, 0) W 25
(1)
E = (0, 0, 0) (0, 0, 0) W 33 (0, 0, 0) (0, 0, 0)

W 41 (0, 0, 0) W 43 (0, 0, 0) W 45
(0, 0, 0) W 52 (0, 0, 0) W 54 (0, 0, 0)

C ( t +1 )
=f ( C ( t )
E ) (
, C 0 )
= In n (2)

Where Inxn denotes the fuzzy identity matrix with TFN, E = [W


] is n n fuzzy
ij
weight matrix with TFNs, which gathers the values of causal edge weights between
concept C i and C j , C (t +1) is the fuzzy state matrix with TFN at certain iteration (t+1),
C (0) is the fuzzy initial matrix with TFN and C (t) is the fuzzy state matrix with TFN
at certain iteration (t), and f is a fuzzy threshold transformation function with TFN.
There are different fuzzy threshold transformation functions that have been used by
researchers [6, 7], [10].
Once the linguistic FCM has been created, it is run directly using Eq. (2). In each
step of cycling, the values of the concepts change according to Eq. (2). After very
little iteration, it will reach to fuzzy steady state (equilibrium state) if one of the fol-
lowing cases occurs [10]:

A fuzzy fixed point attractor: This case is reached when the FCM state vector
remains fixed for successive iterations, for example, 1 2 3 3 3
where the fuzzy vector 3 is known as the fuzzy fixed point attractor.
A fuzzy limit cycle: A sequence of FCM state vector keeps repeating forming a
cycle, for example, 1 2 3 4 5 3 4 5 where the three
fuzzy vector 3 4 5 forming a cycle is known as the fuzzy limit cycle.

3 Linguistic Fuzzy Decision Network (LFDN) Method

LFDN method is proposed for solving MCDM problem to address the uncertainty
situations, an extension of FDM. It calculates the global fuzzy weights among criteria.
The LFDN process to derive criteria priorities can be summarized as follows [6, 7]:
500 B.M. Elomda et al.

Step 1: Calculate the Local Fuzzy Weight (LFW) vector ( L ).


In this step, the Linguistic/Fuzzy Pair-wise Comparison Matrix (LPCM) among crite-
ria is constructed. Therefore, Eq. (3) is applied to derive the LFW. LFW represents
the fuzzy priority weights of the criteria without considering any dependence or feed-
back among criterion. Finally, the gained LFW is normalized.
n n n n
L i = aij = lij , mij , uij , i = 1,...., n (3)

j=1 j=1 j=1 j=1

Step 2: Calculate the fuzzy influence weight matrix among criteria


In this step, the FCM with linguistic values is depicted, and transformed into linguis-
tic adjacency matrix as Eq. (1). The linguistic adjacency matrix is converted to TFN
using Table1. Thus, the fuzzy updating equation as given in Eq. (2) is applied to de-
rive the fuzzy steady-state matrix ( C * ). C * represents the fuzzy causal relationship
among criteria. Finally, the obtained fuzzy steady-state matrix is normalized.
Step 3: Calculate the Global Fuzzy Weight (GFW) vector.
In this step, the fuzzy weighting equation as in Eq. (4) is used to derive GFW. GFW
represents the priority weights of the criteria by considering dependence and feedback
among criterion. Finally, the obtained GFW ( G )is normalized.
The highest priority of criteria is obtained by ranking GFW. Therefore, the Center
Of Area (COA) defuzzification method as in Eq. (5) is applied to convert a vector of
TFNs such as GFW to crisp value.

G L
= L + C (4)
n n n

= (l + m + u ) 3
D(x) (5)

Where ( L n ) is the normalization of the LFW and ( C n ) is the normalization of the


fuzzy steady-state matrix.

4 Multi-Level LFDN Hierarchy Structure (ML-LFDN) Model


The LFDN is a method for solving MCDM problems in fuzzy environment which
have dependent and feedback. Unfortunately, it cant deal with multi-level hierar-
chical structure. Therefore, LFDN method is developed to address the need for
consideration the multi-level hierarchical structure case that consists of goal, criteria,
sub- criteria, sub sub-criteria, etc. down to the lowest level i.e., alternatives level.
ML-LFDN process can be represented as follows:

Step 1: Identify and clarify the nature of the problem which will be solved
Step 2: Build a hierarchy model of elements
This step includes constructing the ML-LFDN model. A ML-LFDN contains of an
overall goal, a criteria/ factors group that based on the goal, a group of alterna-
tives/options that based on the alternatives for reaching the goal. The criteria can be
A Multi-Level Linguistic Fuzzy Decision Network Hierarchical Structure Model 501

further divided down into sub-criteria, sub-sub criteria, and so on, for any number of
levels as the problem requires. After the hierarchy structure is created completely as
in Fig. 2, the overall weights among elements (criteria, sub-criteria, and alternative) at
each level of hierarchy will be derived as shown in step 3, step 4, step 5, step 6 and
step 7 respectively. Fig. 2 illustrates ML-LFDN Hierarchical Structure model.

Objective Goal

Dependent & feedback


Criteria
C1 Ci Cn

Sub- Criteria C11 C1k Ci1 Cij Cn1 Cnm

Alternatives A1 Ai An

Fig. 2. ML-LFDN Hierarchical Structure model

Step 3: Calculate the GFW for criteria in the criteria level


The following steps are executed to derive the GFW:

DMs will be constructed linguistic pairwise comparisons matrix (LPCM) at criteria


level of the hierarchy regarding to their relative significance with first/Goal level.
LPCM in the ML-LFDN assume that the DM can compare any two elements Ci, Cn
at the same level of the hierarchy and provide a linguistic value aij for the ratio of
their importance using table 2. Table 2 illustrates the linguistic preference scale for
ML-LFDN. Once LPCM among criteria have been determined by using a scale i.e.,
LPCM among (C1,.., Ci, , Cn) w.r.t goal. Then, the LFW is derived using the
fuzzy eigenvalue method as in Eq. (3).
Once the FCM is drawn, DM uses table 1 to indicate the influence degree among
criteria. Then, the fuzzy influence among criteria is calculated using Eq. (2)
with fuzzy logistic transformation function as in Eq. (6). Therefore, the FGW for
criteria is obtained using Eq. (4).

( )
= 1 1 + e -xl ,1 1 + e-x m ,1 1 + e-x u ,
f (x) x = ( x l , x m , x u ) (6)
502 B.M. Elomda et al.

Table 2. Linguistic preference scale for pair-wise comparison


fuzzy Definition of fuzzy values TFN scale TFN reciprocal
scale scale
1 Equally Preferred (EP) (1,1,1) (1,1, 1)

2 Equally to Weakly Preferred (EWP) (1, 2, 3) (1/3, 1/2 ,1 )

3 Weakly Preferred (WP) (1,3,5) (1/5, 1/3 ,1 )

4 Weakly to Moderately Preferred (WMP) (2, 4, 6) (1/6, 1/4 ,1/2 )

5 Moderately Preferred (MP) (3,5,7) (1/7, 1/5,1/3)

6 Moderately to strongly Preferred (MSP) (4, 6, 8) (1/8, 1/6 ,1/4 )

7 Strongly Preferred (SP) (5,7,9) (1/9,1/ 7,1/5)

8 Strongly to very strongly Preferred (SVSP) (6, 8, 9) (1/9, 1/8 ,1/6 )

9 Very strongly Preferred (VSP) (7,9,9) (1/9,1/ 9,1/7)


Very strongly to extremely Preferred (VSXP) (8,10,11) (1/11,1/ 10,1/8)
Extremely Preferred (XP) (9,11,11) (1/11,1/ 11,1/9)

Step 4: Calculate the FGW for each group of sub-criteria in sub-criteria level
In this step, the GFW for all sub-criteria can be derived as follows:
DM will compare each group of the sub-criteria w.r.t its parent criteria in the high-
er/criteria level i.e., a set of LPCM is constructed, where each element in a criteria
level is used to compare its children elements/ sub-criteria w.r.t it. This process
will be continuous until all elements in sub-criteria level are compared. Once all
LPCM in sub-criteria level are created i.e., LPCM among (C11,.., C1k) w.r.t C1 ,
LPCM among (Ci1,.., Cij) w.r.t Ci, and until LPCM among (Cn1,.., Cnm) w.r.t Cn, the
LFW for each group of sub- criteria will be obtained using the fuzzy eigenvalue
method as in Eq. (3).
DM will depict a FCM for each group of the sub-criteria having dependent and
feedback i.e., depict a FCM among elements (C11,.., C1k), and for each group of
sub-criteria having dependent and feedback till for (Cn1,.., Cnm). Then, FCM with
linguistic values is applied as in Eq. (2), with fuzzy logistic transformation function
as in Eq. (6) for each one of FCM to get the influence among sub-criteria. Thus,
the FGW will be obtained for each group of the sub-criteria using Eq. (4). While,
for groups of sub-criteria that havent dependent and feedback among them, the
FCM step is ignored in the sub criteria level groups. For example group (Ci1,.., Cij)
in Fig. 2 havent dependent and feedback. In this case, the GFW is becomes equal
to LFW.

Step 5: Calculate the final fuzzy weight for sub-criteria in sub-criteria level
GFW for criteria (Ci) is multiplied by GFW of its sub-criteria (Cij) in the subsequent
lower level to get final fuzzy weight vector for sub-criteria () as follows:

= [i ]T , i=1,............,n (7)

= GFW(Ci ) * ij i = 1,......, n (8)


A Multi-Level Linguistic Fuzzy Decision Network Hierarchical Structure Model 503

i = [GFW(Ci ) * i1 , GFW(Ci ) * i2 ,....,GFW(Ci ) * ij ,......,GFW(Ci ) * im ] (9)

ij = GFW(Cij ) j = (1,......, m ) (10)

Where i is the final fuzzy weight value of criteria (Ci), ij is the final fuzzy weight
value of sub criteria (Cij), and is the final fuzzy weight vector of sub criteria.
Step 6: Calculate the fuzzy weight for alternatives
In this step, the LFW for alternatives can be derived as follows:
DM will be performed LPCM among alternatives w.r.t each sub-criteria in the
higher level i.e., construct a set of LPCM among alternatives, where each element in a
sub-criteria level is used to compare the elements/alternatives immediately below
w.r.t it. For example, concerning the importance w.r.t elements (C11), we constructed
a n x n LPCM containing our comparison element (A1,.., Ai,.., An). This process will
be continuous till all alternatives are compared regarding to each sub-criteria. Once all
LPCM among alternatives are created, the fuzzy eigenvalue method will be used to
derive the LFW for the alternatives with regard to each sub- criteria.
Step 7: Calculate the final fuzzy weight for alternative w.r.t its parent
In the final step, the fuzzy weight matrix as in Eq. (11) consists of the obtained LFW
vectors for the alternatives w.r.t each sub-criterion. Then, the final fuzzy weight vec-
tor for alternative ( ) can be derived as given in Eq. (12).
C11 C1k Ci1 Cij Cn1 Cnm

A1
(A1 )
(A1 ) (A1 ) (A1 ) (A1 ) (A1 )


= Ai
(Ai )
(Ai ) (Ai ) (Ai ) (Ai ) (Ai )

(11)
(A )
An (A n )
(A n )
(A n ) (A n )
(A n )
n

= (12)

Finally, the final fuzzy weight vector for alternative ( ) is ranked using COA as in
Eq. (5) to determine the alternative with highest priority.

5 Case Study

In order to test the developed model, it has been used to solve a real life problem in
agricultural to choose the best crop among available crops for cultivation in their
fields in Egypt. Agricultural domain experts are asked to use the proposed model to
solve their complex problem as follows:
Step 1: Identify and clarify the nature of the problem which will be solved
Agricultural expert tries to cultivate a crop according to the following four criteria:
Temperature, Water, Marketing and Soil for choosing the best alternative from the
following crops Wheat, Corn, Rice, and Fababean.
504 B.M. Elomda et al.

Step 2: Build a hierarchy model of elements


Fig. 3 shows the ML-LFDN hierarchical structure model for the crop selection prob-
lem. The goal of our problem in selecting for cultivated crop in Egypt is identified in
the first level. The second/criteria level contains: Temperature, Water, Soil and Mar-
keting. The third level (or the lowest level) of the hierarchy contains of the alterna-
tives, which are the different types of crop to be evaluated in order to select the best
crop.
Step 3: Calculate the GFW for criteria in the criteria level
In this step, the expert compares the importance among criteria using the fuzzy im-
portance scale, given in table 2 to compare any two criteria by the question "How
important or strongly". Therefore, the LPCM among criteria is given in table 3. Table
3 is converted to table 4 by substituting the corresponding TFNs for each linguistic
value and reciprocal TFNs using table 2. Table 4 shows pairwise comparison matrix
among criteria with TFNs. Therefore, the local fuzzy weights vector for each criterion
is derived using Eq. (5). So, the local fuzzy weight vector ( L ) is:

L = ((0.311,0.486,0.797), (0.6, 0.941, 1.538), (0.65, 1, 1.538), (0.066, 0.084, 0.123)) T


The expert draws the FCM with linguistic values to indicate the influence among
criteria. The fuzzy relationships between concepts are described using a degree of
influence. Expert describes this degree of influence using table 1. Therefore, The
FCM for the considered case study is shown in Fig. 4. The adjacency matrix obtained
from FCM with linguistic values ( E ) is shown in table 5. The fuzzy adjacency ma-
trix is transformed to the adjacency matrix with TFNs using table 1. Table 6 illustrates
the adjacency matrix with TFNs. The fuzzy updating equation as Eq. (2) is applied to
obtain the fuzzy steady state-matrix. Table 7 shows the fuzzy steady-state matrix us-
ing fuzzy logistic transformation function. Therefore, we can derive the global fuzzy
weights ( G ) for criteria in criteria level as:
= ( (0.679,1.154,2.014), (0.956,1.593,2.72), (1.016,1.668,2.738),(0.397,0.397,1.205) )
T
G(Criteria)

Thus, defuzzifiy the GFW vector to get the ranking among criteria using Eq. (5) as:
) = (1.282, 1.756, 1.807, 0.666) T
D (Gn

It is found that, the selection criterion with highest priority is Water, while the cri-
terion with lowest priority is marketing.
Step 4: Calculate the FGW for each group of sub-criteria in sub-criteria level
Step 5: Calculate the final fuzzy weight for sub-criteria in sub-criteria level
Since there is no sub-criteria level in the original definition of the problem, then there
is no need to go through these steps. Therefore we are going to step 6.

Step 6: Calculate the fuzzy weight for alternatives


In this step, the expert compares the importance among alternatives w.r.t the corre-
sponding elements in the higher level (i.e., for each criterion in criteria level), obtain-
ing a LPCM. For Example, in terms of Temperature pairwise comparisons, determine
A Multi-Level Linguistic Fuzzy Decision Network Hierarchical Structure Model 505

the preference of each alternative over another. Table 8 shows LPCM among alterna-
tives w.r.t Temperature criteria. LPCM is converted to pairwise comparison matrix
with TFN as in table 9 using table 2. Also, table 10, table 11, and table 12, shows
linguistic pairwise comparison matrix among alternative w.r.t criteria Soil, Water and
Marketing respectively. Now, Eq. (3) is applied to get the LFW for alternative w.r.t
parent criteria in the upper level. So, the obtained LFW for the alternatives w.r.t. each
criterion is given as in Eq. (13). So, the ranking for all alternatives w.r.t. Temperature,
Soil, Water and Marketing respectively using COA method are shown in table 13.
Thus, from tables 13, it is clear that, the selection criterion with highest priority is
Corn/ Rice w.r.t Temperature, Rice w.r.t Soil, Wheat w.r.t Water, and Rice w.r.t
Marketing respectively.

Step 7: Calculate the final fuzzy weight for alternative w.r.t its parent
In this step, the final fuzzy weight vector ( ) for alternative is derived using Eq.
(12). So, the fuzzy weights vector is ranked using COA method as in Eq. (5). Table
14 illustrates the final ranking of the obtained fuzzy weight vector for alternative.
Thus, from tables 14 it is found that the selection criterion with highest priority is
Rice, while the criterion with lowest priority is Fababean.

Level 1: Goal Select a crop

Level 2: Criteria

Temperature Soil Water Marketing


Alternatives

-Wheat -Wheat -Wheat -Wheat


-Corn -Corn -Corn -Corn
-Rice -Rice -Rice -Rice
-Fababean -Fababean -Fababean -Fababean

Fig. 3. ML-LFDN hierarchical structure model for crop selection problem

Table 3. LPCM among criteria w.r.t goal


Temperature Soil Water Marketing
Temperature EP SP
Soil SP EP EP SP
Water SVSP EP SP
Marketing EP

Table 4. Pairwise comparison matrix with TFNs


Temperature Soil Water Marketing
Temperature (1, 1, 1) (1/9,1/ 7,1/5) (1/9, 1/8 ,1/6 ) (5,7,9)
Soil (5,7,9) (1, 1, 1) (1,1,1) (5,7,9)
Water (6, 8, 9) (1, 1, 1) (1, 1, 1) (5,7,9)
Marketing (1/9,1/ 7,1/5) (1/9,1/ 7,1/5) (1/9,1/ 7,1/5) (1, 1, 1)
506 B.M. Elomda et al.

Soil
Strong

Weak Medium strong


Temperature Water
Strong
Weak
Medium Weak
Marketing

Fig. 4. FCM with linguistic values among criteria for crop selection

Table 5. Adjacency matrix with linguistic values ( E ) obtained from FCM in Fig. 4
Temperature Soil Water Marketing
Temperature No influence Weak Strong Medium
Soil No influence No influence Medium Strong Weak
Water No influence Strong No influence weak
Marketing No influence No influence No influence No influence

Table 6. Adjacency matrix with TFNs ( E )


Temperature Soil Water Marketing
Temperature (0,0,0) (0,0,0.25) (0.75,1,1)) (0.25,0.5,0.75)
Soil (0,0,0) (0,0,0) (0.5,0.75,1) (0,0,0.25)
Water (0,0,0) (0.75,1,1) (0,0,0) (0,0,0.25)
Marketing (0,0,0) (0,0,0) (0,0,0) (0,0,0)

* ) using fuzzy logistic transformation function


Table 7. Fuzzy steady-state matrix ( C
Temp. Soil Water Marketing
Temp. (0.5,0.5,0.5) (0.647,0.708,0.783) (0.809,0.884,0.907)) (0.593,0.679,0.824)
Soil (0.5,0.5,0.5) (0.640,0.702,0.737) (0.768,0.855,0.903) (0.531,0.562,0.738)
Water (0.5,0.5,0.5) (0.779,0.853,0.872) (0.682,0.758,0.798) (0.531,562,0.739)
Marketing (0.5,0.5,0.5) (0.622,0.675,0.709) (0.665,0.732,0.770) (0.531,0.562,0.678)

Table 8. LPCM among alternatives w.r.t Temperature criteria


Wheat Corn Rice Fababean
Wheat EP EP EP
Corn EP EP MP
Rice MP EP EP
Fababean EP

Table 9. Pairwise comparison matrix with TFNs among alternatives w.r.t Temperature
Wheat Corn Rice Fababean
Wheat (1,1,1) (1,1,1) (1/3, 1/5,1/7) (1,1,1)
Corn (1,1,1) (1,1,1) (1,1,1) (3,5,7)
Rice (3,5,7) (1,1,1) (1,1,1) (1,1,1)
Fababean (1,1,1) (1/3, 1/5,1/7) (1,1,1) (1,1,1)
A Multi-Level Linguistic Fuzzy Decision Network Hierarchical Structure Model 507

Table 10. LPCM among alternatives w.r.t Soil criteria


Wheat Corn Rice Fababean
Wheat EP EP MSP
Corn SP EP
Rice MSP EP MSP
Fababean SP EP

Table 11. LPCM among alternatives w.r.t Water criteria


Wheat Corn Rice Fababean
Wheat EP SP SP EP
Corn EP SP EP
Rice EP
Fababean MSP EP

Table 12. LPCM among alternatives w.r.t. Marketing criteria


Wheat Corn Rice Fababean
Wheat EP SVSP MSP
Corn EP MSP
Rice SVSP SP EP SP
Fababean EP

Temperature Soil Water Marketing

Wheat (0.314,0.4,0.555) (0.339,0.582,1.02) (0.6,1,1.667) (0.399,0.658,1.069)



= Corn (0.6,1,1.666) (0.346,0.594,1.045) (0.356,0.571,0.933) (0.187,0.316,0.551) (13)


Rice (0.6,1,1.666) (0.5555,1,1.8) (0.067,0.091,0.138) (0.6071,1,1.6470)

Fababean (0.314,0.4,0.555) (0.347,0.595,1.05) (0.35,0.563,0.917) (0.049,0.064,0.1)

Table 13. The ranking of the crisp weight vector for alternatives w.r.t each criterion

Crisp Temperature Soil Water Marketing


Weights COA COA COA COA
Ranking Ranking Ranking Ranking
Value Value Value Value
Wheat 0.423 2 0.647 4 1.089 1 0.708 2
Corn 1.089 1 0.662 3 0.620 2 0.351 3
Rice 1.089 1 1.118 1 0.099 4 1.085 1
Fababean 0.423 2 0.664 2 0.609 3 0.071 4

Table 14. The ranking of the final fuzzy weight vector ( ) for alternative
The final weight ( ) Fuzzy value COA value Ranking
Wheat (1.305, 3.317, 9.744) 4.789 2
Corn (1.174, 3.178, 9.417) 4.589 3
Rice (1.248, 3.296, 10.612) 5.052 1
Fababean (0.921, 2.373, 6.605) 3.299 4
508 B.M. Elomda et al.

6 Conclusion

Linguistic Fuzzy Decision Network (LFDN) method has been proposed to overcome
the drawback of Fuzzy Decision Map (FDM) method. However, it can't rank the ac-
tions/alternative to select the appropriate alternative. So, in this paper, we propose an
improved LFDN method, called Multi-Level LFDN hierarchical structure (ML-
LFDN) model. The proposed ML-LFDN model is examined through a crop selection
case study, which cannot be solved using the original LFDN model. The obtained
results ensured the effectiveness of ML-LFDN model. As a future work, make com-
plex problem are planned to be solved using the proposed model.

References
1. Zhenghai, A.: A new TOPSIS with triangular fuzzy number and uncertain weight based on
cosines similar degree. In: Eighth IEEE International Conference on Computational Intel-
ligence and Security (CIS), China, November 17-18, pp. 1721 (2012)
2. Zhang, S.-F., Liu, S.-Y., Zhai, R.-H.: An extended GRA method for MCDM with interval-
valued triangular fuzzy assessments and unknown weights. Computers & Industrial Engi-
neering 61, 13361341 (2011)
3. Saaty, T.L.: Decision making with dependence and feedback: The analytic network pro-
cess. RWS Publications, Pittsburgh (1996)
4. Saaty, T.L.: The analytic hierarchy process. McGraw-Hill, New York (1980)
5. Yu, R., Tzeng, G.H.: A soft computing method for multi-criteria decision making with de-
pendence and feedback. Applied Mathematics and Computation 180, 6375 (2006)
6. Elomda, B.M., Hefny, H.A., Hassan, H.A.: An extension of fuzzy decision maps for multi-
criteria decision-making. Egyptian Informatics Journal 14, 147155 (2013)
7. Elomda, B.M., Hefny, H.A., Hassan, H.A.: MCDM method based on improved fuzzy deci-
sion map. In: IEEE International Conference on Electronics, Circuits, and Systems (ICECS
2013), Abu Dhabi, UAE, December 8-11, pp. 225228 (2013)
8. Kosko, B.: Fuzzy cognitive maps. International Journal on ManMachine Studies 24(1),
6575 (1986)
9. Axelrod, R.: Structure of decision: the cognitive maps of political elites. Princeton Univer-
sity Press, Princeton (1976)
10. Elomda, B.M., Hefny, H.A., Hassan, H.A.: Fuzzy cognitive map with linguistic values. In:
IEEE International Conference on Engineering and Technology (ICET 2014), Cairo,
Egypt, April 19-20 (2014)

You might also like