Sie sind auf Seite 1von 12

ARTICLE IN PRESS

Int. J. Production Economics 122 (2009) 176187

Contents lists available at ScienceDirect

Int. J. Production Economics


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

Development of a process mining system for supporting knowledge


discovery in a supply chain network
H.C.W. Lau , G.T.S. Ho, Y. Zhao, N.S.H. Chung
Department of Industrial and Systems Engineering, The Hong Kong Polytechnic University, Hunghom, Kowloon, Hong Kong

a r t i c l e i n f o

abstract

Available online 31 May 2009

In todays competitive environment, business organizations are forced to maintain their


competitive advantage by their ability to cut costs, increase revenue and uncover hidden
issues. In order to enhance the visibility and transparency of value added information in
a supply chain network, a process mining system is proposed for discovering a set of
fuzzy association rules based on the daily captured logistics operation data, within the
network. The proposed methodology provides all levels of employees with the ability
to enhance their knowledge and understanding of the current business environment.
Once interesting association rules have been extracted, organizations can identify the
root-causes of quality problems in a supply chain and improve performance by netuning the conguration of process parameters in specied processes. The application of
the proposed methodology in a case company has also been studied. The prototype
system has been developed and evaluated after performing a spatial analysis.
The results obtained indicate that the system is capable of extracting high-quality and
actionable information in the case company.
& 2009 Elsevier B.V. All rights reserved.

Keywords:
Supply chain network
Business intelligence
Customer satisfaction

1. Introduction
Due to a high degree of uncertainty in customer
demands and complicated supply chain architecture,
there has been a signicant growth in the employment
of supply chain management concepts to improve the
level of customers satisfaction. It is well known that with
the supply chain concept, achievement of customer
satication is not the responsibility of one department
or functional business sector, but the duty of the whole
supply chain from the upsteam to the downstram
members. The recognition of this fact is considered one
of the critial elements of success for an enterprise. On the
other hand, the everchanging economic situation and
erce competition in todays globalized business environment requires the enterprise to be more cautious about

 Corresponding author. Tel.: +852 2766 6628; fax: +852 2362 5267.

E-mail address: mfhenry@inet.polyu.edu.hk (H.C.W. Lau).


0925-5273/$ - see front matter & 2009 Elsevier B.V. All rights reserved.
doi:10.1016/j.ijpe.2009.05.014

customer satisfaction. Thus, the challenging task for


companies today is to globally investigate the key factors
that affect the level of satisfaction of the end customers
within the supply chain. Although a number of empirical
and scientic techniques have been developed which
consider customer satiscation, previous research has not
addressed this issue thoroughly enough.
Ghodsypour and OBrien (2001) focused on the
problem of supplier selection in the case of multiple
suppliers. It aimed at minimizing the total cost of logistics,
including net price, storage, transportation and ordering
costs but it did not consider the possible effects of cost
minimization on the end subject (customers). Traditionally, various individual business sectors/processes are
concerned solely with their own specic performance in
the supply chain and pay little attention to their
contribution to the satiscation of the end customer.
Recently, Lee et al. (2007) employed lateral transshipment between the units of the given echelon so as
to improve the service level (i.e. satisfaction). Hill and

ARTICLE IN PRESS
H.C.W. Lau et al. / Int. J. Production Economics 122 (2009) 176187

Pakkala (2007) proposed a base stock inventory policy for


an immediate dispatch of multi-item products. Note that
the extra stocks for lateral trans-shipment are not always
available since each sector tries to reduce the inventory.
Moreover, lateral trans-shipment is much difcult
to ensure double wins on both sectors, which limits
its application in practice. Some literatures discussed the
implementation or enhancement of customer satisfaction
zkan,
in other industrial processes (Turkyilmaz and O
2007; Chien et al., 2002). They usually laid special
emphasis on the concepts rather than approaches.
Signicantly, Yadav and Geol (2008) employ the customer
satisfaction as the ultimate target to improve product
quality for the automotive industry in USA.
This research is concerned with the introduction
of data mining approach, which aims at discovering the
hidden relationships in a supply chain network among
daily operational data. It then provides an overview
of how to make improvement or take corrective actions
in order to achieve customer satisfaction. The last stage is
to implement the changes, provided that the generalized
combination of process parameters is approved by project
managers. The proposed methodology extracts and
identies the interrelationship between the congurations of process variables within the supply chain workow, with the objective of delivering high-quality outputs

177

from smooth running processes to produce customer


satisfaction.
This paper is divided into three main sections. The rst
section is a literature review about the current practice
of supply chain management and the application of the
association rule mining approach adopted in different
industries. Section 2 describes the proposed methodology.
The case problem scenario is dened in section three. The
nal section draws a conclusion from the application of
the proposed methodology in the case and the results
achieved are discussed.

2. Literature review
Because of the complex, stochastic, dynamic nature
and multi-criteria of the logistics operations along the
supply chain, the existing approaches cannot design,
operate and evaluate an agile supply chain (SC) effectively
(Jain and Deshmukh, 2008; Wang and Cao, 2008).
Managing a supply network is not easy where different
sources of uncertainty and bullwhip effect between
various entities exist in the SC (Bayraktar et al., 2008).
The evolution of information technology has helped the
logistics companies to become more efcient in using
various resources in order to satisfy customers needs.

Suppliers

Manufacturing

Process Mining system


(PMS)
Extract IF-THEN fuzzy
association rule

Quality assurance

Supply chain manager


Centralized database

Packaging
Inventory

...

Distribution centre 1

Distribution centre 2

Distribution centre i

Process flow
...

Data flow
Customer 1

Customer 2

Fig. 1. The framework of a process mining system in a supply chain network.

Customer j

ARTICLE IN PRESS
178

H.C.W. Lau et al. / Int. J. Production Economics 122 (2009) 176187

Table 1
List of notations used in the proposed i-PM algorithm.
n

The number of integrated supply chain workow


records
N f1; 2; . . . ; ng
The set of index of integrated supply chain workow
records
The ith integrated supply chain workow records,
Wi
8i 2 N
s
The number of departments in an integrated supply
chain workow record
S f1; 2; . . . ; sg
The set of index of departments in a supply chain
workow record
The ath department, 8a 2 S
Da
m
The number of processes in a given supply chain
workow record
M f1; 2; . . . ; mg
The set of index for processes in a supply chain
workow record
P aj
The jth process in ath department, 8a 2 S and 8j 2 M
The number of relevant process parameters in jth
kaj
process of ath department
K aj f1; 2; . . . ; kaj g The set of index for relevant process parameters in
jth process of ath department
Q ajt
The tth relevant process parameter for P aj, 8t 2 K aj
The number of fuzzy regions for Q ajt
hajt
Hajt f1; 2; . . . ; hajt g The set of index for fuzzy regions for Q ajt
Rajtl
The lth fuzzy region of Q ajt , 8l 2 Hajt
PV iajt
The quantitative value of Q ajt of P aj of Da in W i
The fuzzy set converted from PV iajt
f iajt
f iajtl
count ajtl
max  count ajt
max  Rajt

aajt
l
Cr
Lr

The membership value of Q ajt of P aj of Da of W i in


region Rajtl
The summation of f iajtl values where 8i 2 N
The maximum count value among count ajtl values,
8l 2 Hajt
The fuzzy region of Q ajt with max  count ajt
The predened minimum support threshold of Q ajt ,
8a 2 S, 8j 2 M, 8t 2 K aj
The predened minimum condence threshold
The set of candidate itemsets with r items
The set of large itemsets with r items

Data Mining Association rules are being widely used to


support industrial and logistics applications nowadays.
Different intelligent logistics information systems are
implemented to help provide decision support in material
sourcing, production scheduling and physical distribution
systems (Ketikidis et al., 2008). Studies indicate that there
are many ways of developing the data mining tools that
support knowledge discovery within the supply chain
network. Tsai et al. (2009) developed an association
clustering technique to mine the correlated demands
so as to conduct a joint replenishment policy, which
signicantly reduce the operation cost. Ha and Krishnan
(2008) proposed a hybrid approach incorporating multiple techniques to select competitive suppliers in a supply
chain. Chen and Wu (2005) adopted the association-based
clustering approach to discover the associations between
orders, such as when the occurrence of some orders in a
batch may cause the occurrence of other orders in the
same batch. In fact, mining association rules from the data
warehouse has recently become one of the most effective
methods for extracting hidden knowledge. Basically, an
association rule indicates a relationship among items,
such as the occurrence of certain other items in the same
transaction (Agrawal et al., 1993).

Early research of association rule discovery primarily


concentrated on boolean association rules, which merely
determine whether an item is present in a transaction or
not (Agrawal and Srikant, 1994). Researchers then started
to pay attention to quantitative properties of association
rules since numerical attributes usually contain much
more plentiful information with quantitative meaning.
However, the disadvantage of quantitative attributions is
the sharp boundary that strictly differentiates the elements near the boundary (Kaya and Alhajj, 2005).
Obviously, it is not consistent with the human perception. To address this problem, fuzzy sets by introducing
gradual membership have been widely employed in
mining quantitative association rules. More signicantly,
the fuzzy set theory has been widely introduced in
intelligent systems as it allows uncertainty and imprecision in the behaviors of data mining (Lee et al., 2008; Chen
and Chang, 2008). Based on such consideration, Jain and
Deshmukh (2008) proposed a method based on fuzzy
association rule mining to support the evaluation of the
agility in supply chains. The rules and knowledge patterns
discovered are used to provide better decision support,
which forms the basis of support for streamlining the
integrated logistics operations.
There are a number of information systems, which aim
to streamline integrated logistics operations. However,
there have been relatively few attempts at the information
systems for conducting data analysis within the whole
supply chain network. The association rule mining is
separately and independently employed to the certain
echelon. Liao et al. (2008) investigated association rules
for product line and branch extension in the retailer level.
Wang (2008) developed the soft association rule mining
using relaxed fuzzy quantier for item promotions.
It can be observed that the above related works pay
attention to the association rule mining that three
decisions were separately applied in past studies. Actually
subjects in the supply chain network are closely related
to each other and tiny changes in these subjects may
generate a signicant difference in the other mined rules
(Kim, 2007). If one pays much attention to a specic level
(or area) one may fail to obtain the effective/correct
association rules in a whole supply chain network as the
attributes or items of the associate rule also interact with
elements in other levels of the supply chain. In this
research, the process mining system with an iterative
process mining (i-PM) algorithm embedded for supporting knowledge discovery is proposed for use in the basic
supply chain workow. This i-PM algorithm was evaluated
in a slider manufacturing company to show how it can
nd the relationships between process parameters and
the presence of quality problems (Lau et al., 2009).

3. Framework of a process mining system in a supply


chain network
This paper proposes a process mining system equipped
with the i-PM algorithm to extract hidden information
related to customer satisfaction based on daily captured
logistics operation data. Fig. 1 presents a typical process

ARTICLE IN PRESS
H.C.W. Lau et al. / Int. J. Production Economics 122 (2009) 176187

179

Fig. 2. Flowchart of the i-PM algorithm for association rules discovery in the supply chain network (Lau et al., 2009).

mining system framework for supporting knowledge


discovery from the historical process data captured in
the purchasing department, the manufacturing department and the transportation department. The daily
process data among the above functional departments is
collected and stored in the centralized data warehouse.
The customer satisfaction report is also collected and
transferred into the data warehouse after the delivery
of the nished products. The proposed process mining
system can analyze correlations between combinations
of process parameters, and the degree of customer
satisfaction by using the proposed iterative process

mining algorithm. The managers from different functional


units can then obtain the association rules for assisting
them to make strategic decisions to achieve customer
satisfaction.

3.1. Application of proposed i-PM algorithm in supply chain


network
The objective of i-PM algorithm is to identify the
improvement actions for supply chain network optimization in term of fuzzy rules. This algorithm adopts the

ARTICLE IN PRESS
180

H.C.W. Lau et al. / Int. J. Production Economics 122 (2009) 176187

fuzzy generalized mining algorithm proposed by Hong


et al. (2003), which integrates data mining technologies
with fuzzy sets, to identify valuable association rules from
quantitative process information. The developed approach
has successfully applied to discover association rules for
the quality enhancement in a semiconductor manufacturer (Lau et al., 2009). In this algorithm a lower-bound
support threshold is the critical parameter that directly
determines the quality of association rules. In real
applications, the values of minimum support for various
process parameters are required to be adjusted or tuned in
a exible way. Therefore, the strategy proposed by Lee
et al. (2005), i.e. multiple lower-bound supports using
maximum constraints, is adopted. The mechanism of the
proposed i-PM algorithm is briey described below.
The notations in this algorithm are tabulated as follows
(Lau et al., 2009) (Table 1).
Data engineers record a set of n supply chain workow
records. Each integrated workow record consists of s
functional departments and each department consists
of m processes. Each process (jth) consists of different
numbers of process parameters kaj . A set of minimum
support value aajt , a predened condence value l and a
set of membership functions are predened by relevant
experts. From these inputs, a set of mined fuzzy association rules which identify the relationships between
process parameters and customer satisfaction will be
identied through the mechanism, as shown in Fig. 2.

Table 2
The symbols of process parameters in a supply chain network.
Process parameters

Symbol

Purchasing
department

Production
department
Logistics
department

Lead time of the suppliers (020 days)

Size of material ordered (5002000 pieces) B


Wavelength of the machine (nm)
C
Fixture angle of the machine (1)
Number of vehicles available for delivery
(521 vehicles)
Number of products per delivery (200500
pieces)

Customer satisfaction factors


Delivery time (016 days)
Number of defective items (132 pieces)

D
E
F

G
H

4. Case scenario
A case scenario is given to illustrate how the proposed
i-PM algorithm extracts generalized fuzzy association
rules from quantitative process data within the supply
chain network. The i-PM algorithm was applied and
evaluated in the manufacturing industry to identify
the root causes of the failure of products, based on the
historical process data. The results indicate that the
algorithm can extract the actionable knowledge effectively (Lau et al., 2009). Different symbols are used to
represent various process parameters and nished quality
features as shown in Table 2. The data set including the six
workow records and each record contains the settings of
process parameters in different processes from three
functional departments are shown in Table 3. Fuzzy
membership values are produced for all items in the
supply chain workow based on the predened membership functions as shown in Fig. 3. The minimum support
for each process parameter is predened and shown in
Table 4. Moreover the condence value l is set at 0.7 and
acts as the threshold for interesting association rules.
Below are the steps of the proposed i-PM algorithm.
Step 1: Transform the quantitative values of the process
parameters and customer satisfaction features in each
supply chain workow record into a fuzzy set using the
membership functions given above. Take the second
process parameter in the second workow record as an
example. As shown in Fig. 4, the crisp value 1000 of
process parameter B is converted into the fuzzy set
which is calculated as (0.33/medium+0.67/low). This step
is repeated for all items in six workow records and the
result is given in Table 5. The converted structure of
process parameters with fuzzy regions is represented as
department.process.process_parameter.fuzzy_region.
Step 2: Calculate the count of each fuzzy region of each
process parameter in the workow record and put the
process parameters in a set C 1 . The process parameter in
C 1 is then called an itemset. Take process parameter A as
an example. Since process parameter A has three fuzzy
regions: low, medium and high, the low fuzzy
region is chosen to demonstrate the calculation of the
count. The count for the low fuzzy region of process
parameter A is calculated from the six workow records
by adding the fuzzy count of D1 :P1 :A:Low of each record,
and is calculated as (0.83+0+0+0.67+0+0) 1.5. The

Table 3
The expanded supply chain workow record.
Supply chain workow
ID (SID)

1
2
3
4
5
6

Department 1 (purchasing
department) (D1)

Department 2 (production
department) (D2)

Department 3 (logistics
department) (D3)

Process 1 (P1)

Process 2 (P2)

Process 1 (P1)

Process 2 (P2)

Process 1 (P1)

Process 2 (P2)

A
5
8
12
6
8
12

B
800
1000
1500
1100
1100
2000

C
2.5
3
3.5
5.5
5
5

D
620
625
620
628
628
630

E
15
15
5
9
18
18

F
350
300
420
500
350
440

Customer
satisfaction (Q)

G
3
6
8
2
4
8

H
2
1
19
13
7
19

ARTICLE IN PRESS
H.C.W. Lau et al. / Int. J. Production Economics 122 (2009) 176187

Medium

Short

Long

Medium

Low

High

Membership
value

Membership
value

181

0
0

10

15

875

500

20

Parameter A
Low

Medium

1250

1625

2000

Size of material ordered (pieces)

Lead time (days)

Parameter B

High

Very High

5.1

6.2

Low

Medium

High

Membership
value

Membership
value

0
2.65

3.75

600

607

Parameter C
Medium

High

300

380

650

Parameter D
Very High

Medium

Short

1
1

Long

Membership
value

Membership
value

Low

625
Wavelength (nm)

Angle ()

0
200 236

440

500

12

17

21

No. of products per delivery

No. of vehicle available for delivery

Parameter E

Parameter F

Short

Long

Very Long

Low

Medium

High

Very High

19

27

Membership
value

Membership
value

Medium

6
10
Delivery time (days)

13

16

13

32

No. of defect items

Parameter G

Parameter H

Fig. 3. The membership functions of process parameters and customer satisfaction features in a case example.

Table 4
The minimum support of each process parameter.
Process parameter

Min. Support aajt

1.8

1.5

1.5

2.5

2.3

1.9

counts for the fuzzy regions of other process parameters


follow the same procedure and are shown in Table 6.
Step 3and 4: Find the maximum count, max  countajt ,
for each process parameter and the fuzzy region,
max  Rajt , is the region with the maximum counts of a
process parameter. For process parameter A, the counts

ARTICLE IN PRESS
182

H.C.W. Lau et al. / Int. J. Production Economics 122 (2009) 176187

Medium

Low

High

Membership value

0.67
(Low)
0.33
(Medium)

0
500

875 1000
1250
1625
Size of material ordered (pieces)

2000

Fig. 4. Fuzzy set conversion of process parameter B.

for the three fuzzy regions, which are low, medium


and high, are 1.5, 3.04 and 1.46, respectively. Since the
medium fuzzy region has the highest count, max 
count ajt is set as 3.04 and max  Rajt is set as MEDIUM for
process parameter A. The maximum counts and the
corresponding fuzzy regions for each process parameter
are shown below in Table 7.
Step 5: Check whether the value max  countajt of a
region max  Rajt of a process parameter is equal to or
larger than its predened minimum support value aajt . If
the value of a region in Table 7 is equal to or greater than
its minimum support value, it will be selected and put in
the large 1-itemset (L1). Since the counts of the fuzzy
regions with maximum counts of all process parameters
which are shown in Table 7 above are larger than its
minimum support which is shown in Table 4 3, all the
process parameters are then put in the large 1-itemset L1
which is shown in Table 8.
Step 6: The candidate set C2 is generated from L1, and
the supports of the two items in each itemset in C2 must
be larger than or equal to the maximum of their
predened minimum support values. For example, the
minimum support values for process parameter A and D
are 1.8 and 2.0, respectively, and the maximum of
minimum support value of process parameter A and D is
then taken as 2.0. Since the counts of D1 :P 1 :A:Medium and
D2 :P 2 :D:Medium are 3.04 and 4.92, respectively, which is
larger than their maximum of minimum support value
(2.0), itemsets fD1 :P1 :A:Medium; D2 :P 2 :D:Mediumg is put
into the set of candidate 2-itemsets. On the other hand
for the other possible 2-itemsets fD2 :P 1 :C:High; D3 :P 2 :F:
Mediumg, since the count for D2 :P 1 :C:High is 2.49 which is
smaller than their maximum of minimum support value
(2.5). fD2 :P1 :C:High; D3 :P 2 :F:Mediumg is not going to be put
into C2. The other candidates in 2-itemsets are generated
in the same way. The 2-itemsets are shown below in
Table 9.
Step 7: For each newly formed candidate 2-itemsets
in C2
(a) Calculate the fuzzy value and the count of each
candidate 2-itemsets in the workow record. The
count for each 2-itemsets is calculated as the mini-

mum of counts for each item in the workow record.


Take fD1 :P 1 :A:Medium; D2 :P2 :D:Mediumg as an example. The count is calculated from each workow
record. Since fD1 :P 1 :A:Medium; D2 :P2 :D:Mediumg occurs in record 1 concurrently, the count is equal to
0.17 for record 1. Summing up the counts for the 6
records is the count for the 2-itemsets fD1 :P 1 :A:
Medium; D2 :P 2 :D:Mediumg which is calculated as (0.17+
0.67+0.6+0.33+0.67+0.6) 3.04. Other 2-itemsets follow the same way to calculate their count. The result
is shown in Table 10.
(b) The support value of each candidate 2-itemsets is then
calculated and compared with the maximum of the
minimum support values of the items contained in C2.
The item with count equal to or larger than the
predened minimum support value is put into the
large 2-itemsets L2. For fD1 :P1 :A:Medium; D2 :P2 :D:
Mediumg, the minimum support of process parameter
A is 1.8 and D is 2.0. The support for the 2-itemsets
fD1 :P 1 :A:Medium; D2 :P 2 :D:Mediumg is then taken as
the maximum of these two values which is 2.0.
The support values for the 2-itemsets is calculated
and shown in Table 11.

The count of each 2-itemsets, which is shown in Table 10,


is then compared with its corresponding minimum
support value which is shown in Table 11. Still take
fD1 :P1 :A:Medium; D2 :P 2 :D:Mediumg as an example. The
count for fD1 :P1 :A:Medium; D2 :P 2 :D:Mediumg is 3.04 and
the minimum support for it is 2.0. Since the count of 3.04
is larger than its minimum support, it is put into the
large 2-itemsets L2. While for fD1 :P 1 :A:Medium; D3 :P 2 :F:
VeryHighg, its count (1.33) is smaller than its corresponding minimum support (2.5), so it is not going to put into
the large 2-itemsets L2. The rest of the 2-itemsets follows
the same procedure. The result is shown in Table 12.
Step 8: As L2 is not null, go on to the next step.
Step 9: Set r 2; STEPs 69 are repeated and shown in
STEPs 1012.
Step 10: The candidate set C3 is generated from L2, and
the supports of the two items in each itemset in C3 must
be larger than or equal to the maximum of their

ARTICLE IN PRESS









0:4
0:6
1
0:07
0:93
0:2
0:8
1
1
0:5
0:5
1

D1 :P 1 :A:High D1 :P1 :A:Medium


D1 :P2 :B:High
D2 :P 1 :C:Medium D2 :P 1 :C:High
D2 :P 2 :D:High D2 :P 2 :D:Medium
D3 :P 1 :E:High
D3 :P 2 :F:VeryHigh
Q :G:Medium Q :G:High
Q :H:High
6



0:84
0:16

D2 :P2 :D:Medium D2 :P 2 :D:High



0:63
0:37

D2 :P 1 :C:High D2 :P1 :C:VeryHigh



0:6
0:4

D1 :P 2 :B:Medium D1 :P2 :B:Low



0:33
0:67

D1 :P1 :A:Medium D1 :P 1 :A:Low









0:67
0:33
0:6
0:4
0:07
0:93
0:84
0:16
1
0:625
0:375
0:5
0:5
0:75
0:25

D1 :P 1 :A:Medium D1 :P 1 :A:High
D1 :P2 :B:Medium D1 :P2 :B:Low
D2 :P 1 :C:Medium D2 :P1 :C:High
D2 :P 2 :D:Medium D2 :P2 :D:High
D3 :P 1 :E:High
D3 :P 2 :F:High D3 :P 2 :F:Medium
Q :G:Low Q :G:Medium
Q :H:Low Q :H:Medium





0:75
0:25
1
1
1

D3 :P1 :E:Low D3 :P1 :E:Medium


D3 :P 2 :F:VeryHigh
Q :G:Low
Q :H:Medium









0:4
0:6
0:67
0:33
0:77
0:23
0:28
0:72
0:67
0:23
0:5
0:5
1
1

D1 :P 1 :A:High D1 :P 1 :A:Medium
D1 :P2 :B:High D1 :P2 :B:Medium
D2 :P 1 :C:Medium D2 :P 1 :C:Low
D2 :P2 :D:Low D2 :P 2 :D:Medium D3 :P1 :E:Low D3 :P2 :F:VeryHigh D3 :P2 :F:High
Q :G:Medium Q :G:High
Q :H:High









0:17
0:83
1
1
0:28
0:72
0:6
0:4
0:625
0:375
0:25
0:75
1

D1 :P1 :A:Mediu D1 :P 1 :A:Low


D1 :P 2 :B:Low
D2 :P 1 :C:Low
D2 :P 2 :D:Low D2 :P 2 :D:Medium
D3 :P 1 :E:High D3 :P 1 :E:Medium
D3 :P2 :F:High D3 :P 2 :F:Medium
Q :G:Medium Q :G:Low
Q :H:Low









0:67
0:33
0:33
0:67
0:32
0:68
1
0:6
0:4
1
1
1

D1 :P1 :A:Medium D1 :P 1 :A:High


D1 :P2 :B:Medium D1 :P 2 :B:Low
D2 :P 1 :C:Low D2 :P1 :C:Medium
D2 :P2 :D:Medium
D3 :P1 :E:High D3 :P 1 :E:Medium
D3 :P 2 :F:Medium
Q :G:Medium
Q :H:Low

Quantitative values of process variables using fuzzy sets


WID

Table 5
The fuzzy sets transformed from the process data.

H.C.W. Lau et al. / Int. J. Production Economics 122 (2009) 176187

183

predened minimum support values. Take fD1 :P1 :


A:Medium; D2 :P 2 :D:Medium; D3 :P 1 :E:Highg as an example.
The minimum support value is taken as the maximum
of the minimum support of fD1 :P1 :A:Medium; D2 :P 2 :D:
Mediumg, fD1 :P 1 :A:Medium; D3 :P 1 :E:Highg, and fD2 :P 2 :D:
Medium; D3 :P 1 :E:Highg which are all 2.0 and thus
the maximum count, 2.0, is taken as the count for
fD1 :P1 :A:Medium; D2 :P 2 :D:Medium; D3 :P1 :E:Highg. For fD1 :
P 1 :A:Medium; D2 :P2 :D:Medium; D3 :P 1 :E:Highg to be included
in C3, the counts of all of its two itemsets including fD1 :P 1 :A:Medium; D2 :P 2 :D:Mediumg, fD1 :P1 :A:Medium;
D3 :P1 :E:Highg, and fD2 :P 2 :D:Medium; D3 :P1 :E:Highg which
are 3.04, 2.04 an 2.84, respectively, must be equal to
or larger than the minimum support which is 2.0 in this
case. So fD1 :P1 :A:Medium; D2 :P 2 :D:Medium; D3 :P1 :E:Highg is
put in C3. The candidate 3-itemsets are then found as
fD1 :P1 :A:Medium; D2 :P 2 :D:Medium; D3 :P1 :E:Highg, fD1 :P 1 :A:
Medium; D2 :P 2 :D:Medium; Q :G:Mediumg, fD1 :P 1 :B:Low; D2 :
P 2 :D:Medium; Q :H:Lowg.
Step 11: For each newly formed 3-itemsets C3:
(a) Calculate the fuzzy value and the count of each
3-itemsets in C3: Take fD1 :P1 :A:Medium; D2 :P 2 :D:
Medium; D3 :P1 :E:Highg as an example. The count for
record 3 and 4 is 0 since these three items do
not occur concurrently. While for record 2 is 0.6
which is the minimum among D1 :P1 :A:Medium,
D2 :P 2 :D:Medium, D3 :P 1 :E:High. Similarly, the count for
record 1, 5 and 6 are equal to 0.17, 0.67 and 0.6,
respectively. Thus the count of fD1 :P1 :A:Medium;
D2 :P 2 :D:Medium; D3 :P1 :E:Highg is the summation of
the counts for 6 records, (0.17+0.6+0+0+0.67+0.6)
2.04. The counts of the other 3-itemsets are shown in
Table 13.
(b) Check the count of each itemset in C3 with its
predened minimum support value. Put it in L3 if its
count is equal to or greater than its predened
minimum support. The support value is calculated
as the maximum of their minimum support. For
the
fD1 :P 1 :A:Medium; D2 :P2 :D:Medium; D3 :P 1 :E:Highg,
minimum support for process parameter A, D and E
are 1.8, 2.0, 2.0, respectively. Thus the support for the
3-itemsets
fD1 :P 1 :A:Medium; D2 :P2 :D:Medium; D3 :P 1 :E:Highg
is 2.0. The supports for the 3-itemsets are shown in
Table 14.
Comparing the count of 3-itemsets with its support value,
those itemset with count equal to or larger than the
support value is put in the large 3-itemsets L3.
Since only fD1 :P1 :A:Medium; D2 :P 2 :D:Medium; D3 :P1 :E:Highg
and fD1 : P 1 :A:Medium; D2 :P 2 :D:Medium; Q :G:Mediumg have
counts higher than their support value, they are put into
the large 3-itemsets L3 which is shown in Table 15.
Step 12: Since L3 is not null, the previous steps are
repeated. No candidate 4-itemset, C4, is generated and L4
is null.
Step 13: The association rules for each large q-itemsets, q4 2, are constructed by forming all possible
association rules and by calculating the condence values

ARTICLE IN PRESS
184

H.C.W. Lau et al. / Int. J. Production Economics 122 (2009) 176187

Table 6
The fuzzy counts of the itemsets in C1.

Table 9
Items in C2.

Process items

Count

Process items

D1 :P1 :A:Low
D1 :P1 :A:Medium
D1 :P1 :A:High
D1 :P2 :B:Low
D1 :P2 :B:Medium
D1 :P2 :B:High
D2 :P1 :C:Low
D2 :P1 :C:Medium
D2 :P1 :C:High
D2 :P1 :C:VeryHigh
D2 :P2 :D:Low
D2 :P2 :D:Medium
D2 :P2 :D:High
D3 :P1 :E:Low
D3 :P1 :E:Medium
D3 :P1 :E:High
D3 :P2 :F:Medium
D3 :P2 :F:High
D3 :P2 :F:VeryHigh 2
Q.G. Low
Q.G. Medium
Q.G. High
Q.H. Low
Q.H. Medium
Q.H. High

1.5
3.04
1.46
2.47
1.86
1.67
1.55
1.59
2.49
0.37
0.56
4.92
0.52
1.75
1.05
3.2
1.75
1.48
2.67
2.25
2.75
1
2.75
1.25
2

fD1 :P 1 :A:Medium; D1 :P 2 :B:Lowg


fD1 :P 1 :A:Medium; D2 :P 1 :C:Highg
fD1 :P 1 :A:Medium; D2 :P 2 :D:Mediumg
fD1 :P 1 :A:Medium; D3 :P 1 :E:Highg
fD1 :P 1 :A:Medium; D3 :P 2 :F:VeryHighg
fD1 :P 1 :A:Medium; Q :G:Mediumg
fD1 :P 1 :A:Medium; Q :H:Lowg
fD1 :P 2 :B:Low; D2 :P 1 :C:Highg
fD1 :P 2 :B:Low; D2 :P 2 :D:Mediumg
fD1 :P 2 :B:Low; D3 :P 1 :E:Highg
fD1 :P 2 :B:Low; Q :G:Mediumg
fD1 :P 2 :B:Low; Q :H:Lowg 2
fD2 :P 1 :C:High; D2 :P 2 :D:Mediumg
fD2 :P 1 :C:High; D3 :P 1 :E:Highg
fD2 :P 1 :C:High; Q :G:Mediumg 2
fD2 :P 1 :C:High; Q :H:Lowg 2
fD2 :P 2 :D:Medium; D3 :P 1 :E:Highg 2
fD2 :P 2 :D:Medium; D3 :P 2 :F:VeryHighg
fD2 :P 2 :D:Medium; Q :G:Mediumg
fD2 :P 2 :D:Medium; Q :H:Lowg
fD3 :P 1 :E:High; D3 :P 2 :F:VeryHighg
fD3 :P 1 :E:High; Q :G:Mediumg
fD3 :P 1 :E:High; Q :H:Lowg
fD3 :P 2 :F:Medium; Q :G:Mediumg
fD3 :P 2 :F:Medium; Q :H:Lowg

Table 7
The maximum fuzzy counts and the corresponding fuzzy regions of the
itemsets in C1.
Process items

Count

D1 :P1 :A:Medium
D1 :P2 :B:Low
D2 :P1 :C:High
D2 :P2 :D:Medium
D3 :P1 :E:High
D3 :P2 :F:VeryHigh 2
Q.G. Medium
Q.H. Low

3.04
2.47
2.49
4.92
3.2
2.67
2.75
2.75

Table 8
Items in large 1-itemset L1.
Process items

Count

D1 :P1 :A:Medium
D1 :P2 :B:Low
D2 :P1 :C:High
D2 :P2 :D:Medium
D3 :P1 :E:High
D3 :P2 :F:VeryHigh 2
Q.G. Medium
Q.H. Low

3.04
2.47
2.49
4.92
3.2
2.67
2.75
2.75

of all association rules. Take IF fD1 :P 1 :A:Medium; D2 :P 2 :


D:Mediumg, THEN fD3 :P 1 :E:Highg as an example. The
condence value of this rule is calculated as
D1 :P 1 :A:Medium \ D2 :P 2 :D:Medium \ D3 :P 1 :E:High
D1 :P 1 :A:Medium \ D2 :P 2 :D:Medium
2:04

0:67
3:04
Results for all association rules are shown below in Table 16.

Table 10
The fuzzy counts of the itemset in C2.
Process items

Count

fD1 :P 1 :A:Medium; D1 :P 2 :B:Lowg


fD1 :P 1 :A:Medium; D2 :P 1 :C:Highg
fD1 :P 1 :A:Medium; D2 :P 2 :D:Mediumg
fD1 :P 1 :A:Medium; D3 :P 1 :E:Highg
fD1 :P 1 :A:Medium; D3 :P 2 :F:VeryHighg
fD1 :P 1 :A:Medium; Q :G:Mediumg
fD1 :P 1 :A:Medium; Q :H:Lowg
fD1 :P 2 :B:Low; D2 :P 1 :C:Highg
fD1 :P 2 :B:Low; D2 :P 2 :D:Mediumg
fD1 :P 2 :B:Low; D3 :P 1 :E:Highg
fD1 :P 2 :B:Low; Q :G:Mediumg
fD1 :P 2 :B:Low; Q :H:Lowg
fD2 :P 1 :C:High; D2 :P 2 :D:Mediumg
fD2 :P 1 :C:High; D3 :P 1 :E:Highg
fD2 :P 1 :C:High; Q :G:Mediumg
fD2 :P 1 :C:High; Q :H:Lowg
fD2 :P 2 :D:Medium; D3 :P 1 :E:Highg2
fD2 :P 2 :D:Medium; D3 :P 2 :F:VeryHighg
fD2 :P 2 :D:Medium; Q :G:Mediumg
fD2 :P 2 :D:Medium; Q :H:Lowg
fD3 :P 1 :E:High; D3 :P 2 :F:VeryHighg
fD3 :P 1 :E:High; Q :G:Mediumg
fD3 :P 1 :E:High; Q :H:Lowg
fD3 :P 2 :F:Medium; Q :G:Mediumg
fD3 :P 2 :F:Medium; Q :H:Lowg

1.57
1.6
3.04
2.04
1.33
2.34
1.51
0.8
2.19
1.76
1.32
2.07
2.27
1.86
1
0.5
2.84
2.31
2.75
2.15
1
1.85
1.95
1.6
1.75

Step 14: Compare the condence values of the rules


with the predened condence threshold which is 0.7 in
this case. Table 17 gives the rules with condence
values equal to or larger than the predened condence
threshold.
Step 15: Keep the rules with nished department/
process variables in the consequent part only, and output

ARTICLE IN PRESS
H.C.W. Lau et al. / Int. J. Production Economics 122 (2009) 176187

Table 11
The support of the 2-itemsets.

185

Table 15
Items in large 3-itemsets L3.

Process items

Count

Process items

Count

fD1 :P 1 :A:Medium; D1 :P2 :B:Lowg


fD1 :P 1 :A:Medium; D2 :P1 :C:Highg
fD1 :P 1 :A:Medium; D2 :P2 :D:Mediumg
fD1 :P 1 :A:Medium; D3 :P1 :E:Highg
fD1 :P 1 :A:Medium; D3 :P2 :F:VeryHighg
fD1 :P 1 :A:Medium; Q :G:Mediumg
fD1 :P 1 :A:Medium; Q :H:Lowg
fD1 :P 2 :B:Low; D2 :P 1 :C:Highg
fD1 :P 2 :B:Low; D2 :P 2 :D:Mediumg
fD1 :P 2 :B:Low; D3 :P 1 :E:Highg
fD1 :P 2 :B:Low; Q :G:Mediumg
fD1 :P 2 :B:Low; Q :H:Lowg
fD2 :P 1 :C:High; D2 :P 2 :D:Mediumg
fD2 :P 1 :C:High; D3 :P 1 :E:Highg
fD2 :P 1 :C:High; Q :G:Mediumg
fD2 :P 1 :C:High; Q :H:Lowg
fD2 :P 2 :D:Medium; D3 :P1 :E:Highg2
fD2 :P 2 :D:Medium; D3 :P2 :F:VeryHighg
fD2 :P 2 :D:Medium; Q :G:Mediumg
fD2 :P 2 :D:Medium; Q :H:Lowg
fD3 :P 1 :E:High; D3 :P2 :F:VeryHighg
fD3 :P 1 :E:High; Q :G:Mediumg
fD3 :P 1 :E:High; Q :H:Lowg
fD3 :P 2 :F:Medium; Q :G:Mediumg
fD3 :P 2 :F:Medium; Q :H:Lowg

1.8
1.8
2.0
2.0
2.5
2.3
1.9
1.5
2.0
2.0
2.3
1.8
2.0
2.0
2.3
1.9
2.0
2.5
2.3
2.0
2.5
2.5
2.5
2.5
2.5

fD1 :P 1 :A:Medium; D2 :P2 :D:Medium; D3 :P 1 :E:Highg


fD1 :P 1 :A:Medium; D2 :P2 :D:Medium; Q :G:Mediumg

2.04
2.34

Table 12
Items in large 2-itemsets L2.
Process items

Count

fD1 :P 1 :A:Medium; D2 :P2 :D:Mediumg


fD1 :P 1 :A:Medium; D3 :P1 :E:Highg
fD1 :P 1 :A:Medium; Q :G:Mediumg
fD1 :P 2 :B:Low; D2 :P 2 :D:Mediumg
fD1 :P 2 :B:Low; Q :H:Lowg
fD2 :P 1 :C:High; D2 :P 2 :D:Mediumg
fD2 :P 2 :D:Medium; D3 :P1 :E:Highg2
fD2 :P 2 :D:Medium; Q :G:Mediumg
fD2 :P 2 :D:Medium; Q :H:Lowg

3.04
2.04
2.34
2.19
2.07
2.27
2.84
2.75
2.15

Table 13
The fuzzy counts of the itemsets in C3.
Process items

Count

fD1 :P 1 :A:Medium; D2 :P2 :D:Medium; D3 :P 1 :E:Highg


fD1 :P 1 :A:Medium; D2 :P2 :D:Medium; Q :G:Mediumg
fD1 :P 2 :B:Low; D2 :P 2 :D:Medium; Q :H:Lowg

2.04
2.34
1.79

Table 16
All possible association rules and their corresponding condence values.
Association rules

Condence
value

If {D1 :P 1 :A:Medium, D2 :P 2 :D:Medium}, then


{D3 :P 1 :E:High}
If {D1 :P 1 :A:Medium, D3 :P 1 :E:High} then
{D2 :P 2 :D:Medium}
If {D3 :P 1 :E:High, D2 :P 2 :D:Medium} then
{D1 :P 1 :A:Medium}
If {D1 :P 1 :A:Medium} then {D3 :P 1 :E:High,
D2 :P 2 :D:Medium}
If {D2 :P 2 :D:Medium} then {D1 :P 1 :A:Medium,
D3 :P 1 :E:High}
If {D3 :P 1 :E:High} then {D1 :P 1 :A:Medium,
D2 :P 2 :D:Medium}
If {D1 :P 1 :A:Medium, D2 :P 2 :D:Medium, } then
{Q :G:Medium}
If {D1 :P 1 :A:Medium , Q :G:Medium} then
{D2 :P 2 :D:Medium}
If {D2 :P 2 :D:Medium, Q :G:Medium then {D1 :P 1 :A:Medium}
If {D1 :P 1 :A:Medium} then {D2 :P 2 :D:Medium,
Q :G:Medium}
If {D2 :P 2 :D:Medium} then {D1 :P 1 :A:Medium,
Q :G:Medium}
If {Q :G:Medium} then {D1 :P 1 :A:Medium,
D2 :P 2 :D:Medium}
If {D1 :P 1 :A:Medium} then {D2 :P 2 :D:Medium}
If {D2 :P 2 :D:Medium} then {D1 :P 1 :A:Medium}
If {D1 :P 1 :A:Medium} then {D3 :P 1 :E:High}
If {D3 :P 1 :E:High} then {D1 :P 1 :A:Medium}
If {D1 :P 1 :A:Medium} then {Q :G:Medium}
If {Q :G:Medium} then {D1 :P 1 :A:Medium}
If {D1 :P 2 :B:Low} then {D2 :P 2 :D:Medium}
If {D2 :P 2 :D:Medium} then {D1 :P 2 :B:Low}
If {D1 :P 2 :B:Low} then {Q :H:Low}
If {Q :H:Low} then {D1 :P 2 :B:Low}
If {D2 :P 1 :C:High} then {D2 :P2 :D:Medium}
If {D2 :P 2 :D:Medium} then {D2 :P 1 :C:High}
If {D2 :P 2 :D:Medium} then {D3 :P 1 :E:High}
If {D3 :P 1 :E:High} then {D2 :P 2 :D:Medium}
If {D2 :P 2 :D:Medium} then {Q :G:Medium}
If {Q :G:Medium} then {D2 :P 2 :D:Medium}
If {D2 :P 2 :D:Medium} then {Q :G:Medium}
If {Q :G:Medium} then {D2 :P 2 :D:Medium}

0.67
1
0.72
0.67
0.41
0.6
0.77
1
0.85
0.77
0.48
0.85
1
0.62
0.67
0.64
0.77
0.85
0.89
0.45
0.84
0.75
0.91
0.46
0.58
0.89
0.56
1
0.44
0.78

Table 14
The support of the 3-itemsets.
Process items

Support

fD1 :P 1 :A:Medium; D2 :P2 :D:Medium; D3 :P 1 :E:Highg


fD1 :P 1 :A:Medium; D2 :P2 :D:Medium; Q :G:Mediumg
fD1 :P 2 :B:Low; D2 :P 2 :D:Medium; Q :H:Lowg2

2.0
2.3
2.0

the rules to users as interesting association rules as


presented in Table 18.
Finally, three fuzzy association rules are extracted
based on the ve supply chain workow records in

this case example. The rules indicate that three process


parameters (suppliers lead time, size of material ordered
and xture angle of the machine) are the important
factors which lead to different levels of customer
satisfaction. The second association rule in Table 18
suggests that low material orders bring the low number
of defective items. Meanwhile it also indicates that too
many orders may result in the increasing of defective
items in the case example. This reminds that the company
managers pay attention to the maximal tolerance of

ARTICLE IN PRESS
186

H.C.W. Lau et al. / Int. J. Production Economics 122 (2009) 176187

Table 17
The association rules with condence values 4 0.7.
Fuzzy association rules

Condence
value

If {D1 :P1 :A:Medium, D3 :P 1 :E:High} then


{D2 :P 2 :D:Medium}
If {D3 :P1 :E:High, D2 :P 2 :D:Medium} then
{D1 :P 1 :A:Medium}
If {D1 :P1 :A:Medium, D2 :P 2 :D:Medium, } then
{Q :G:Medium}
If {D1 :P1 :A:Medium, Q :G:Medium} then
{D2 :P 2 :D:Medium}
If {D2 :P2 :D:Medium, Q :G:Medium then {D1 :P 1 :A:Medium}
If {D1 :P1 :A:Medium} then {D2 :P 2 :D:Medium,
Q :G:Medium}
If {Q :G:Medium} then {D1 :P 1 :A:Medium,
D2 :P2 :D:Medium}
If {D1 :P1 :A:Medium} then {D2 :P 2 :D:Medium}
If {D1 :P1 :A:Medium} then {Q :G:Medium}
If {Q :G:Medium} then {D1 :P 1 :A:Medium}
If {D1 :P2 :B:Low} then {D2 :P 2 :D:Medium}
If {D1 :P2 :B:Low} then {Q :H:Low}
If {Q :H:Low} then {D1 :P 2 :B:Low}
If {D2 :P1 :C:High} then {D2 :P 2 :D:Medium}
If {D3 :P1 :E:High} then {D2 :P 2 :D:Medium}
If {Q :G:Medium} then {D2 :P 2 :D:Medium}
If {Q :G:Medium} then {D2 :P 2 :D:Medium}

1
0.72
0.77
1
0.85
0.77
0.85
1
0.77
0.85
0.89
0.84
0.75
0.91
0.89
1
0.78

Table 18
The association rules which satisfy the relevant constraints.
Fuzzy association rules

Condence
value

If {D1 :P1 :A:Medium, D2 :P 2 :D:Medium, } then


{Q :G:Medium}
If {D1 :P2 :B:Low} then {Q :H:Low}
If {D1 :P1 :A:Medium} then {Q :G:Medium}

0.77
0.84
0.77

material orders and accept the moderate size of material


orders. In Table 18 the third associate rule states the lead
time of suppliers is the main factor to affect the delivery
time. If the end customers require prompt delivery, the
selection of suppliers with shorter lead time is the rst
choice in comparison with other processes that expedite
the delivery. As for the rst association rule, it appears to
indicate the xture angle of the machine is a very
important technology during production or a time consuming process. This needs more investigation to conrm
the inuence of this production process parameter on the
delivery time.
This methodology can be applied to a large scale
supply chain network to extract some actionable knowledge to streamline the whole workow. Before introducing the developed system, the analysis engineers in the
supply chain network can quantize the customer satisfaction using their own satisfaction indices. For example,
in this case example, the satisfaction indices are the
delivery time and number of defective items. By using this
methodology, they can identify process parameters closely related to the customer satisfaction and then improve
or optimize these process parameters aiming to enhance

customer satisfaction. They thus can quantize the new


customer satisfaction and also measure the improvements
of the customer satisfaction in comparison with the old
value. Moreover, the measurement of the improvements
of customer satisfaction reects the feasibility and utility
of the application of the i-PM method to the supply chain
network.
5. Discussion
In this research, the i-PM algorithm is proposed and
the focus is on discovering the correlations among huge
amounts of daily captured logistics operational data. This
is done so as to support knowledge discovery along the
integrated logistics workow. This algorithm is designed
to be applied in the cases where a huge amount of data is
stored in the data warehouse. If the enterprise does not
collect data regularly or if the number of records of stored
data is not enough, this algorithm is inappropriate and
may reduce the reliability and quality of mining rule sets.
This may limit the application of the developed system in
practice.
The fuzzy sets derived from knowledge acquisition
used in the proposed system can have a variety of shapes
including triangle, trapezoid, parabolic, etc. In order to
provide adequate representation of expert knowledge and
simplify the process of computation, a triangle shape is
used in representing different fuzzy sets along the
processes within the supply chain. The framework of
i-PM algorithm can be modied in the future by allowing
different shapes of fuzzy sets to be computed. Furthermore, the determination of membership functions and
fuzzy sets by relying on the knowledge of experts is quite
time-consuming and subjective. It is suggested that some
AI techniques with self-learning ability be incorporated to
derive the fuzzy sets automatically.
6. Conclusions
The prototype process mining system for supporting
knowledge discovery in a supply chain network is
proposed and developed. The new methodology of the
proposed i-PM algorithm has been evaluated in a case
study and the algorithm shows its potential to gure out
the primary factors that have a great effect upon the
satisfaction of the end customer in a supply chain. Three
process parameters (suppliers lead time, size of material
ordered and xture angle of the machine) have been
identied as the important factors strongly related to
the level of customer satisfaction in the case example. The
association rules mined by the i-PM algorithm reveal that
in the case study the short lead time of suppliers brings
the fast delivery, and moderate material orders avoid
increasing defective items. Signicantly, in comparison
with numbers of vehicles as well vehicle capacity, the lead
time of suppliers takes the critical action on the delivery
time. We also note that some association rule, such as the
rst rule in Table 18 is relatively difcult to understand its
meaning. It is suggested that people need to integrate
such association rule with practical situation for in-depth

ARTICLE IN PRESS
H.C.W. Lau et al. / Int. J. Production Economics 122 (2009) 176187

analysis or resort to other techniques to examine or even


exclude it. This is a universal challenge for the techniques
of mining association rules to provide effective rules and
avoid dubious rules.
Furthermore, the advantage of the i-PM algorithm is
that it can be used and applied in an enterprise where its
centralized data warehouse is huge. A large number of
itemset useful association rules concerning process parameters and customer satisfaction within a supply chain
network, can be easily extracted. Adequate sample data
can ensure the quality of association rules. However,
the disadvantage is that it may take a large amount of
computational time to identify the signicant association
rules; and the membership functions of different parameters along the logistics workow are always subjective.
Selection of suitable fuzzy sets, like most fuzzy
quantier (Wang, 2008) is one suggestion for the possible
future work. In commercial applications, the developed
process mining system could be based on the integration
of the i-PM algorithm with the right fuzzy sets. Therefore,
another future research is necessary to validate the
performance of the system in a variety of applications
through extensive practical evaluations and measurement
of the improvement on the customer satisfaction.

Acknowledgement
The authors wish to thank the Research Committee of
the Hong Kong Polytechnic University for their support of
this project. Also, many thanks go to the anonymous
reviewers for their useful comments on the previous
version of this paper.
References
Agrawal, R., Srikant, R., 1994. Fast algorithms for mining association
rules. In: Proceedings of International Conference on Very Large
Databases, pp. 487499.
Agrawal, R., Imielinski, T., Swami, A., 1993. Mining association rules
between sets of items in large databases. In: Proceedings of
ACM SIGMOD International Conference on Management of Data,
pp. 207216.
Bayraktar, E., Koh, S.C., Gunasekaran, A., Sari, K., Tatoglu, E., 2008. The
role of forecasting on bullwhip effect for E-SCM applications.
International Journal of Production Economics 113, 193204.
Chen, S.H., Chang, S.M., 2008. Optimization of fuzzy production
inventory model with unrepairable defective products. International
Journal of Production Economics 113, 887894.
Chen, M.-C., Wu, H.-P., 2005. An association-based approach to order
batching considering customer demand patterns. Omega 33, 333343.

187

Chien, T.-k., Su, C.-H., Su, C.-t., 2002. Implementation of a customer


satisfaction program: a case study. Industrial Management and Data
Systems 102, 252259.
Ghodsypour, S.H., OBrien, C., 2001. The total cost of logistics in supplier
selection, under conditions of multiple sourcing, multiple criteria
and capacity constraint. International Journal of Production Economics 73, 1527.
Ha, S.H., Krishnan, R., 2008. A hybrid approach to supplier selection for
the maintenance of a competitive supply chain. Expert Systems with
Applications 34, 13031311.
Hill, R.M., Pakkala, T.P.M., 2007. Base stock inventory policies for a multiitem demand process. International Journal of Production Economics
109, 137148.
Hong, T.-P., Lin, K.-Y., Wang, S.-L., 2003. Fuzzy data mining for interesting
generalized association rules. Journal of Fuzzy Sets and Systems 138,
255269.
Jain, V.B., Deshmukh, S.G., 2008. A new approach for evaluating agility in
supply chains using Fuzzy Association Rules Mining. Engineering
Applications of Articial Intelligence 21, 367385.
Kaya, M., Alhajj, R., 2005. Genetic algorithm based framework for
mining fuzzy association rules. Fuzzy Sets and Systems 152,
587601.
Ketikidis, P.H., Koh, S.C.L., Dimitriadis, N., Gunasekaran, A., Kehajova, M.,
2008. The use of information systems for logistics and supply chain
management in South East Europe: current status and future
direction. Omega 36, 592599.
Kim, S.W., 2007. Organizational structures and the performance of
supply chain management. International Journal of Production
Economics 106, 323345.
Lau, H.C.W., Ho, G.T.S., Chu, K.F., Ho, W., Lee, C.K.M., 2009. Development
of an intelligent quality management system using fuzzy association
rules. Expert Systems with Application 36, 18011815.
Lee, Y.-C., Hong, T.-P., Lin, W.-Y., 2005. Mining association rules with
multiple minimum supports using maximum constraints. International Journal of Approximate Reasoning 40, 4454.
Lee, Y.H., Jung, J.W., Jeon, Y.S., 2007. An effective lateral transshipment
policy to improve service level in the supply chain. International
Journal of Production Economics 106, 115126.
Lee, Y.-C., Hong, T.-P., Wang, T.-C., 2008. Multi-level fuzzy mining with
multiple minimum supports. Expert Systems with Applications 34,
458468.
Liao, S.-H., Chen, C-M., Wu, C.-H., 2008. Mining customer knowledge for
product line and brand extension in retailing. Expert Systems with
Applications 34, 17631776.
zkan, C., 2007. Development of a customer satisfaction
Turkyilmaz, A., O
index model: an application to the Turkish mobile phone sector.
Industrial Management and Data Systems 107, 672687.
Tsai, C.Y., Tsai, C.Y., Huang, P.W., 2009. An association clustering
algorithm for can-order policies in the joint replenishment
problem. International Journal of Production Economics 117,
3041.
Wang, F.H., 2008. On discovery of soft associations with most fuzzy
quantier for item promotion applications. Information Sciences 178,
18481876.
Wang, J., Cao, D.B., 2008. Relationships between two approaches for
planning manufacturing strategy: a strategic approach and a
paradigmatic approach. International Journal of Production Economics 115, 349361.
Yadav, O.P., Geol, P.S., 2008. Customer satisfaction driven quality
improvement target planning for product development in automotive industry. International Journal of Production Economics 113,
9971011.

Das könnte Ihnen auch gefallen