An effective supplier selection method for constructing

a competitive supply-relationship

Gye Hang Honga,b,*, Sang Chan Parkb, Dong Sik Janga, Hyung Min Rhoc

aIndustrial System and Information Engineering, Korea University, Seoul, South KoreabDepartment of Industrial Engineering, Korea Advanced Institute of Science and Technology, KAIST 373-1 Kusong-dong,

Yusong-ku, Taejon 305-701, South KoreacCADCAM Research Center, Korea Institute of Science and Technology, Seoul, South Korea

Abstract

We propose an effective supplier selection method to maintain a continuous supply-relationship with suppliers. Costs have been sharply

increasing and profit decreasing as the global competition among companies has increased and customer demands have diversified in the

current business environment. Many other functions are now outsourced globally to strengthen competition. As a result, one of the issues is

how to select good suppliers which can maintain a continuous supply-relationship.

We suggest a mathematical programming model that considers the change in suppliers’ supply capabilities and customer needs over a

period in time. We design a model which not only maximizes revenue but also satisfies customer needs. The suggested model is applied to

supplier selection and management of the agriculture industry in Korea.

q 2005 Elsevier Ltd. All rights reserved.

Keywords: Supply selection; Data mining; Mixed-integer programming; Supply-relationship management; Customer demand

1. Introduction

In the paper, we propose an effective supplier selection

method to maintain a continuous supply-relationship with

suppliers. The proposed method is designed to consider

multiple criteria such as quantity, price, quantity and

delivery, and to evaluate the change in suppliers’ supply

capabilities over a period in time. It helps to select the

optimal suppliers which can maximize revenue with

different procurement conditions according to each period

of time satisfied.

Costs have been sharply increasing and profit decreasing

as the global competition among companies has increased

and customer demands have diversified in the current

business environment. Many companies are trying to reduce

their costs while satisfying various customer demands.

0957-4174/$ - see front matter q 2005 Elsevier Ltd. All rights reserved.

doi:10.1016/j.eswa.2004.12.020

* Corresponding author. Address: Department of Industrial Engineering,

Korea Advanced Institute of Science and Technology, KAIST 373-1

Kusong-dong, Yusong-ku, Taejon 305-701, South Korea. Tel.: C82 42 869

2960; fax: C82 42 869 3110.

E-mail address: kistduck@kaist.ac.kr (G.H. Hong).

They are strengthening their core competencies and out-

sourcing many other functions globally. They have

organized global supply networks and have competed with

other supply networks (Choy, Lee, & Lo, 2002). One of the

issues is how to select good suppliers which can maintain a

continuous supply-relationship.

There are four research subjects within the research field

of supplier selection: problem definition, formulation of

criteria, pre-qualification, and final selection (de Boer,

Labro, & Morlacchi, 2001), with the latter two being most

actively pursued.

The pre-qualification step can be defined as the process

of reducing the set of ‘all’ suppliers to a smaller set of

acceptable suppliers and of categorizing methods into four

categories: categorical methods, data envelopment analysis

(DEA), clustering analysis (CA), and case-based reasoning

(CBR) system (de Boer et al., 2001).

Holt (1998) reviewed several decisional methods (CA,

bespoke approaches, multi-attributed analysis, multiple

regression and multivariate discriminant analysis) which

are being applied in supplier selection and compared the

methods. He suggested that CA offers the greatest potential

Expert Systems with Applications 28 (2005) 629–639

www.elsevier.com/locate/eswa

G.H. Hong et al. / Expert Systems with Applications 28 (2005) 629–639630

for pre-qualifying all suppliers because it reduces the

probability of rejecting a ‘good’ supplier too early in the

process via subjective reduction of the often large original

set. CA enlarges the scope for rationalization of the

selection process by identifying the criteria involved.

Because of these merits, we use one of the several CA

methods for evaluating all available suppliers in the pre-

qualification step.

The methods suggested in the final selection step are

categorized into the following models: linear weighting,

total cost of ownership, mathematical programming (MP),

statistics, and AI. Most methods belong to linear weighting

and MP models.

MP allows the decision-maker to formulate the decision

problem in terms of a mathematical objective function that

subsequently needs to be maximized (e.g. maximizing

profit) or minimized (e.g. minimizing costs) by varying the

values of variables in the objective function (e.g. the amount

ordered with supplier X). Weber and Desai (1996)

illustrated how a parallel axis analysis can be used to

identify alternative paths in which inefficient vendors can

become efficient providers of a product. Weber, Current,

and Desai (1998) expanded their models to negotiate with

suppliers selected by the MOP model under non-cooperative

negotiation strategy. Especially, they showed that values of

supplier selection changed according to the number of

suppliers.

In linear weighting models weights are given to the

criteria, where the biggest weight indicates the highest

importance. Ratings on the criteria are multiplied by their

weights and summed in order to obtain a single figure for

each supplier. The supplier with the highest overall rating

can then be selected. Lee, Ha, and Kim (2001) suggested the

supplier selection and management system (SSMS) which

used the linear weighting model to calculate the weights of

tangible and intangible criteria and to rank the suppliers’

performance. The characteristic of the system is the process

which identifies the weak criteria of selected suppliers by

comparing with alternative suppliers. SSMS informs us of

the directions for improving supplier performance.

Fig. 1. Changes in supplier’s capability

We have formulated the following conjectures by

carefully examining the existing research results.

(1)

cond

We must evaluate the change in suppliers’ supply

capabilities over a period in time.

(2)

Suppliers should be evaluated with more than one

criterion (e.g. price, quality, delivery performance).

(3)

We should design a multi-steps model for reducing the

complexity of the supplier selection problem.

(4)

We must select suppliers which can maximize the

revenue with different procurement conditions by each

period in time satisfied.

(5)

We identify the changing supply conditions of the

selected suppliers over the period in time for taking

necessary actions.

We suggest an MP model that considers the change in

suppliers’ supply capabilities and customer needs over a

period in time. We design the model which not only

maximizes revenue but does so while satisfying customer

needs.

2. Define problems

2.1. Evaluate the change in suppliers’ supply capabilities

over a period in time

Suppliers’ supply capabilities are changed over time.

Talluri and Sarkis (2002) demonstrated that the perform-

ance of a selected supplier is monitored across 18 time

periods using DEA and that the supplier’s efficiency in each

period in time is changed. It is difficult for suppliers to

maintain the same capability condition during all supply

periods in industries which have seasonal demands and

experience a wide fluctuation of capability condition over

periods in time. As shown in Fig. 1, a supplier supplies the

products which have a higher price than the ideal purchasing

condition in time T. However, this supply condition is

changed to the lower price and lower quality in time TC1,

ition over a period in time.

G.H. Hong et al. / Expert Systems with Applications 28 (2005) 629–639 631

to the lower quantity in time TC2 and to the lower quality

and higher price in time TC3. All suppliers cannot maintain

the same capability conditions during all analyzing periods

because of changes of delivery condition, inventory level

and market environments.

Customer consuming patterns vary over time in indus-

tries producing seasonal products. Companies must adopt

different procurement strategies for each period in time and

evaluate their suppliers. Although suppliers maintain the

same capabilities, they cannot satisfy the changed procure-

ment strategies. Therefore, we consider the change in both

suppliers’ supply capabilities and procurement conditions

over a period in time.

However, most supplier selection models have not

considered the change in supply capabilities and procure-

ment conditions over a period in time. Recent research

works (de Boer et al., 2001; Holt, 1998; Lee et al., 2001;

Weber & Desai, 1996; Weber et al., 1998) consider only the

comprehensive supply capability conditions of the total

analyzing periods to evaluate suppliers. Therefore, the

methods produce a result that does not select the supplier

which is in good standing during parts of all periods because

of lower performance during the other parts. In addition,

they may select suppliers which do not satisfy the new status

of changing consuming patterns after any period of time.

Moreover, they cannot take necessary actions because they

do not know when its performance should be decreased.

Thus, it is important that we divide all analyzing periods

into several meaningful period units (MPUs), evaluate the

supply conditions of each MPU with the identified

procurement condition of each MPU and combine the

results.

2.2. Multi-criteria model

We consider multiple criteria for evaluating the suppli-

ers’ capability conditions. In an early study on supplier

selection criteria, Dickson (1966) identified 23 criteria that

have been considered by purchasing managers in various

Fig. 2. Multiple crit

supplier selection problems. Since the Dickson study, many

researchers have identified that important criteria varied by

industry and buying situations and have suggested multi-

criteria models. These criteria are classified by the approach

as follows.

In his portfolio approach, Kraljic (1983) identified the

purchasing situation in terms of two factors: profit impact

and supply risk. Profit impact includes such elements as the

(expected) monetary volume involved with the good-

s/services to be purchased and the impact on (future)

product quality. Indicators of supply risk may include the

availability of goods/services under consideration and the

number of potential suppliers.

We consider such criteria as price, delivery, quality,

quantity and their variance, as shown in Fig. 2. In the pre-

qualification step, we use the criteria (e.g. delivery,

variance) for evaluating supply risk and the criteria (e.g.

quality, quantity, and price) for profit impact. By applying

these criteria, we can find several sets of suppliers which

have low supply risk and above necessary profit. In the final

selection step, we use the criteria (revenue, customer

satisfaction) for evaluating profit impact and lower risk

while satisfying the customer.

2.3. Optimization model

We must select the suppliers which maximize the

revenue of a purchasing company with the change in

procurement condition over the period in time satisfied.

Current mathematic models focus on optimization of

revenue or cost without considering how important

customer satisfaction is. As mentioned above, consuming

patterns of important customers are changed over time.

These changes have an effect on all nodes (wholesalers,

manufacturers, delivery and so on) in the supply network

and they finally have an effect on procurement condition of

the company.

Therefore, the company must have different procurement

strategies to reflect changing consuming patterns by each

eria by steps.

Fig. 3. Selecting the suppliers which maximize revenue with the procurement condition satisfied.

G.H. Hong et al. / Expert Systems with Applications 28 (2005) 629–639632

period. Under the different procurement conditions by each

period, we should select the optimal suppliers which can

maximize their revenue while satisfying the different

procurement condition by each period.

We solve the problem as shown in Fig. 3. Suppliers are

divided into several groups having similar features of supply

conditions in the pre-qualification step. Each group may

satisfy some criteria of procurement condition in time t but

may not satisfy all the conditions. We maximize the revenue

by assigning more suppliers of the group which is similar to

the procurement condition than those of other groups.

3. System framework

In this paper, we consider a number of suppliers and the

multiple criteria and we evaluate the suppliers’ supply

capabilities of all periods in time. Accordingly, it is very

time-consuming to evaluate all the suppliers and find the

optimal suppliers. We therefore suggest a three-step model

which consists of preparing, pre-qualification, and final

selection.

(1) Preparation step

We decide which criteria are used to evaluate suppliers.

Then, we transform both internal and external company data

into a suitable form. Data are summarized by the unit of

periods in terms of weeks or packs of 10 days.

(2) Pre-qualification step

We select the pre-qualified supplier sets which satisfy the

procurement condition among all available suppliers. For

pre-qualifying suppliers, we divide the total analyzing

periods into several MPUs. An MPU is defined as the period

in which the sales environment can be distinguished from

that of the previous period. Then we identify the procure-

ment condition by each MPU.

We search for potential suppliers from internal and

external company data. We segment potential suppliers into

several groups having similar features. We evaluate the

features of each group to satisfy the procurement condition

by each MPU and select candidate sets by each MPU.

(3) Final selection step

We finally select the optimal suppliers for maximizing

revenue while satisfying the procurement condition by each

MPU. We identify the changing supply condition of the

selected suppliers over the period in time and suggest

management directions for them (Fig. 4).

4. Supplier selection method

In this section, we explain the subordinate processes of

each step specially. We apply supplier selection and

management system to the agriculture industry in Korea

to explain the method. This system helps wholesalers or

food manufacturers to select the optimal suppliers which

can construct a continuous supply-relationship among

suppliers producing fruits in the Korean agriculture

industry.

4.1. Define criteria and summarize data

(1) Define criteria

We define important criteria of both supply risk and

supply profit. In terms of supply risk, we define the criteria

which can evaluate whether or not a supplier supplies the

wanted quantity in the wanted time. Frequency, average

quantity per order, and its variance are defined in the criteria

Fig. 4. System framework.

G.H. Hong et al. / Expert Systems with Applications 28 (2005) 629–639 633

for evaluating supply risk. If a supplier has many transaction

frequencies, average quantity above a certain level and low

variance, they will have a low supply risk. On the other

hand, we define the criteria which can evaluate a profit as

price, quality and quantity.

(2) Summarize data

We transform both internal and external data into the

information for measuring the criteria as shown in Fig. 5.

We summarize total data to comprehensive information of

Fig. 5. Transform data into the infor

the unit period. The unit period is the smallest unit for

analyzing the supplier’s supply condition and it is decided

according to the purchasing cycle time, the quantity of

collected data per period, etc.

We divide the total period of time into unit periods of

10 days and summarize the data belonging to each unit

period to semantic information. We obtain information of

the supplier (id: 33320331) which supplies a product

(id: 60103). The information consists of the frequency

mation for measuring criteria.

Fig. 6. Divide total analyzing periods into MPUs.

G.H. Hong et al. / Expert Systems with Applications 28 (2005) 629–639634

(the number of transaction within unit period), average

quantity per order and its variance, and average price per

order and its variance.

4.2. Pre-qualification

Table 1

Segmenting customers into four clusters in terms of RFM in MPU T3

Recency Frequency Monetary Number

of case

Relative

importance

of sales (%)

Cluster 1 0.66 0.30 0.54 20 51.95

Cluster 2 0.38 0.98 0.72 4 13.68

Cluster 3 0.22 0.17 0.08 46 16.62

Cluster 4 0.76 0.09 0.05 70 17.75

4.2.1. Divide total analyzing periods into several MPUs and

identify procurement conditions by MPUs

A company’s sales quantity changes over a period in

time, as shown in Fig. 5. Quantities of sales are increased or

decreased during any period and they are stable to a lower or

higher level. We can show that the main customer needs are

changed according to sales variations. When the sales

quantity is increasing, buyers need to find suppliers which

can supply as large quantities as possible. However, when

the sales quantity is a continuous lower level, they need to

delivery the wanted quantity at the right time. Therefore, we

divide the total analyzing periods into several MPUs and

identify customer needs by MPUs.

How do we divide the total analyzing periods into

MPUs? Park and Park (2003) suggested a method for

dividing total periods into several periods. They segmented

sales records of total periods with a genetic algorithm and

linear regression model.

As shown in the left of Fig. 6, the trends of the sales

records are changed over the period in time. Intuitively, we

can see that the sales records of total periods are segmented

into the four groups having similar trend. We use Park’s

method for dividing the total period automatically, and then

can obtain four MPUs as shown in the right of Fig. 6. Then,

we summarize the information by each MPU.

After defining MPU and summarizing the information by

each MPU, we identify the main customers who have an

effect on sales quantity by each MPU and their consuming

pattern. We extract the RFM (Recency, Frequency,

Monetary) information from the sales data for identifying

the main customers. The RFM clustering method is one of

the analyzing methods for discovering customer patterns

(Ha & Park, 1998). We define RFM as follows.

–

Recency. The last period of time of purchases made in

each MPU

–

Frequency. The number of purchases made in each MPU

–

Monetary. Amount of money spent during each MPU

We segment customers into groups having a similar

consuming pattern with the Self-Organizing-Map (SOM),

which is one of the clustering tools. SOM is an unsupervised

learning scheme to train the neural network. Unsupervised

learning comprises those techniques for which the resulting

actions or desired outputs for the training sequences are not

known. The network is only told the input vectors, and it

then self-organizes these inputs into categories.

When segmenting customers in each MPU, we can

obtain the results as shown in Table 1. Customers are

divided into four clusters in period T3. Cluster 1 has the

relative importance of sales among the clusters. As shown in

Table 2, the cluster has the characteristics defined as the

frequency above average value, higher price, larger quantity

Table 2

Identifying customer needs by cluster in MPU T3

Quality Frequency Price Quantity

Cluster 1 1st level 24 21,423 105

Cluster 2 1st and 2nd

level

78 14,911 45

Cluster 3 1st–8th level 14 10,873 28

Cluster 4 1st–4th level 8 16,277 20

Average 14.62 15,197 36

G.H. Hong et al. / Expert Systems with Applications 28 (2005) 629–639 635

and higher quality than others. Therefore, we consider the

two controlling factors of both quantity and quality as

having a higher priority than the other factors for satisfying

the main customers in period T3. After analyzing the

customer patterns for all MPUs, we can find important

controlling factors by each MPU and see that the important

controlling factors are changed according to the periods, as

shown in Table 3. An ideal procurement condition for the

company is set to the controlling factors and their values.

4.2.2. Search available supplier

We must find potential suppliers from the outside

database to increase the number of all alternative suppliers

and discover the best solution. Every supplier satisfying the

following condition will be included as an available

candidate

�p:jt K �pi:t Rcijt (1)

where

�p:jt

Tabl

Cons

Fact

Qual

Freq

Price

Quan

average price for which the buyer j purchases a product

in MPU t

�pi:t

average price for which the supplier i sells a product in

MPU t

cijt

cost required when buyer j changes his supplier to

supplier i in MPU t

As shown in formula (1), buyer j can purchase a product

from supplier i when supplier i can give profit to buyer j

above the cost. We find the available suppliers which satisfy

the condition among the outside suppliers.

4.2.3. Segmentation and select candidate groups

We segment suppliers into several groups which have

similar supply conditions with the SOM method. We

evaluate the supply conditions of the groups and pre-qualify

e 3

uming pattern of main customers over periods in time

or MPU

T1 T2

ity 1st–3rd level 5–7th level

uency High Average

Above average Average

tity High Below average

the groups which can satisfy the maximum value (threshold

value) between the average value of all groups and the

company defined value (the minimum qualifying value of a

company). We conduct this process for all MPUs repeatedly

and obtain the pre-qualified supplier groups by MPU.

4.2.4. Normalization

We normalize data to place all values between 0 and 1

because neural networks work best when all input and

output values are between 0 and 1.

4.2.5. Clustering and pre-qualifying

We design the SOM model which has six inputs and nine

outputs. Inputs are the information of supply condition

such as quality, frequency, price and its variance, and

quantity and its variance and outputs are a cluster

number. After obtaining the results shown in Table 4, we

search for the qualified clusters which satisfy the following

conditions:

1.

QualityOZmax {average value of all clusters, company

defined value}

2.

FrequencyOZmax {average value of all clusters,

company defined value}

3.

QuantityOZmax {average value of all clusters, com-

pany defined value}

4.

Frequency and quantity. If a cluster does not satisfy

conditions 2 and 3 but belongs to either case 1 or case 2,

we can select the cluster because we can adjust the

frequency or quantity through negotiation with the

suppliers.

Total quantity per MPU

Z max faverage value of frequency; company

defined value of frequencyg

!max faverage value of quantity; company

defined value of quantityg

Case 1: The value of frequency is larger than its

threshold and the value of quantity is less than

its threshold. However, the value of multiplying

values of frequency by values of quantity

exceeds or equals the total quantity per MPU.

Case 2: The value of frequency is less than its threshold

and the value of quantity is larger than its

threshold. However, the value of multiplying

T3 T4

1st level 1st and 2nd level

Above average Large variance

High price Very high price

Many quantity Large variance

Table 4

Supply conditions of pre-qualified clusters by MPU

MPU Cluster Quality Frequency Price Price_variance Quantity Quantity_variance No. of supplier

T1 1 0.98 0.36 0.72 0.74 0.34 0.33 11

4 0.87 0.38 0.42 0.34 0.37 0.37 3

7 0.97 0.14 0.53 0.04 0.45 0.09 6

Total average 0.74 0.16 0.41 0.30 0.22 0.15 47

T2 2 0.80 0.09 0.32 0.28 0.16 0.09 4

4 0.97 0.31 0.48 0.60 0.38 0.46 15

5 0.75 0.32 0.31 0.52 0.22 0.19 4

Total average 0.76 0.10 0.32 0.25 0.20 0.14 98

T3 2 0.97 0.46 0.68 0.56 0.45 0.67 6

4 0.80 0.13 0.19 0.10 0.31 0.33 1

5 1.0 0.25 0.62 0.22 0.29 0.35 3

Total average 0.79 0.15 0.41 0.16 0.22 0.21 30

T4 2 0.97 0.35 0.39 0.66 0.20 0.19 17

5 0.89 0.16 0.42 0.40 0.11 0.07 11

Total average 0.67 0.11 0.28 0.30 0.12 0.08 89

G.H. Hong et al. / Expert Systems with Applications 28 (2005) 629–639636

values of frequency by values of quantity

exceeds and equals the total quantity per MPU.

5.

Price. Purchasing price is not an important variable in

the pre-qualification step because it has a different level

of price according to quality level. However, it is

important to evaluate the profit effect of candidates in the

final selection step.

6.

Variance of both price and quantity. It is very important

when evaluating the degree of supply risk. We use the

variables to compare candidates.

4.2.6. Identifying the clusters which can satisfy customer

needs among pre-qualified clusters

We found the controlling factors for satisfying the main

customer needs by MPU in the previews process. Because

the other purpose of pre-qualification is identifying the

clusters which can satisfy customer needs among the pre-

qualified clusters, we need to evaluate all pre-qualified

clusters with the controlling factors by MPU.

For example, we evaluate them with the three

controlling factors of frequency, quantity and quality (see

T1 in Table 3) when comparing the clusters pre-qualified in

T1 in Table 4.

(1)

Quality is not an important factor because all clusters

satisfy above the average quality.

(2)

Clusters 1 and 4 are superior to cluster 7 in terms of

frequency and quantity because cluster 7 does not

satisfy above the average frequency.

(3)

Cluster 4 is superior to cluster 1 in terms of price

because the price of cluster 4 is cheaper than that of

cluster 1.

As shown in the above steps, we compare clusters with

controlling factors in the upper level, and then compare

them with the other factors in the lower level. We can

determine the priority of clusters for all MPUs.

4.3. Final selection

4.3.1. Select the optimal suppliers for maximizing revenue

with customer needs satisfied

We select the suppliers which can maximize revenue

with the procurement condition satisfied and assign optimal

order quantities to them.

We design a mixed integer model and find the optimal

solution with the model. The model has the following three

core characteristics:

(1)

Selected suppliers have maintained a supply-relation-

ship for a long time if they do not have a lower

performance than cost for changing supplier.

(2)

The suppliers which satisfy many parts of the ideal

procurement condition are selected more often than

otherwise suppliers.

(3)

The model considers the change in the supplier’s supply

conditions and the ideal procurement conditions over

period in time.

Max Z ZXI

iZ1

XK

kZ1

XT

tZ1

Rirt !xikt KXI

iZ1

XTK1

tZ1

C !cit

KM !XI

iZ1

XT

tZ1

yit (2)

Subject to

XI

iZ1

XK

kZ1

xikt %Dt; for all t (3)

xikt %minðSuit;L

uitÞ!yikt for all i; k; t (4)

xikt RmaxðSlit; L

litÞ!yikt for all i; k; t (5)

XI

iZ1

XK

kZ1

yikt %Nt for all t (6)

G.H. Hong et al. / Expert Systems with Applications 28 (2005) 629–639 637

XK

kZ1

yikt %Nt !yit for all i; t (7)

XI

iZ1

yikt RrkC1;t

XI

iZ1

yiðkC1Þt R/RrkCK;t

XI

iZ1

yiðkCKÞt for all t

(8)

yi;tC1 Kyit Kcit %0 for all i; t (9)

where

xikt

quantity ordered from the supplier i which belongs to

cluster k in period t

Rikt

revenue per unit made from the supplier i which

belongs to cluster k in period t

Dt

purchasing demand in period t

Suit

maximum order quantity available from supplier i in

period t

Slit

minimum order quantity available from supplier i in

period t

Luit

maximum amount of business to be given to supplier

i in period t

Llit

minimum amount of business to be given to supplier

i in period t

Nt

the number of suppliers to be selected in period t

rkCn,t

ratio of number of suppliers selected in cluster kCn

to number of suppliers selected in cluster k in period t

yikt

Z1 if supplier i of cluster k is selected in period t; 0,

otherwise.

yit

Z1 if supplier i is selected in period t; 0, otherwise.

cit

Z1 if supplier i which was selected in the previous

period t is not selected in current period tC1; 0,

otherwise.

C

the cost incurred when one supplier is changed to

another supplier

M

artificial big value; it is used to protect that any

supplier which is not selected in period t is regarded

as selected supplier. For example, we assume that

supplier 1 included in cluster 1 in t1 is not selected

and supplier 1 included in cluster 2 in t1. That is,

y111Z0, y121Z0. We can know that supplier 1 is not

selected in t1 in the case (e.g. y11Z0). However, y11

can have all values of 0 and 1 when it is presented by

Eq. (7). The big M is needed for solving the problem.

Objective function (2) shows the maximization of

revenue with the following conditions satisfied during the

total planning periods.

Constraint (3) shows the purchasing demand in MPU t.

The total order quantity assigned to suppliers cannot exceed

the total quantity of purchasing demand in MPU t.

Constraints (4) and (5) show the supplier’s potential

system constraints and the purchaser’s policy constraints as

described in the model of Weber (1998). That is, suppliers

have the constraints of minimum and maximum supply

quantities and purchasers have the constraints of minimum

and maximum order quantities placed with particular

suppliers. We incorporate these constraints into our model.

Constraint (6) is a limitation of the number of suppliers

which are selected in MPU t. It can be defined as the policy

of the purchaser. The number of suppliers is related to the

supplier risk and the management cost. As the number of

suppliers is increased, so the management cost is increased

and the supply risk is decreased. However, we can set a

small number of suppliers and simultaneously the supply

risk is decreased because of Constraint (8).

Constraint (7) is the expression for evaluating whether or

not a supplier is selected in MPU t. Because a supplier can

belong to several clusters, one supplier can be regarded as

another supplier. To solve this problem, we include the

expression in the model.

Constraint (8) shows a limitation for selecting the

suppliers which satisfy more parts of the customer demands

than other suppliers in MPU t. We evaluate how suppliers of

each cluster satisfy customer demands and give priority to

them in the previous step. Because cluster 4 is superior to

clusters 1 and 7 in the case, we can set that the number of

cluster 4 is r times as large as the number of cluster 1 or the

number of cluster 7.

Constraint (9) is the expression for evaluating whether or

not the supplier selected in MPU t is changed in MPU tC1.

If a supplier is changed in MPU tC1, a changing cost is

incurred in the objective function (2).

We can choose the final suppliers and their order quantity

by MPU with the model and determine their supply

conditions from features of the cluster.

4.3.2. Identify the change in supplier’s supply condition

over period in time

We can determine the following facts by comparing the

final selected suppliers as shown in Table 5.

–

Supplying periods. Supplier 37080000 supplies his

products from MPU T1 to MPU T3 while supplier

33320000 supplies his products during all periods.

–

Management directions by MPU. The supplier’s man-

agement directions are changed according to MPUs.

Supplier 37080000 is managed in terms of variance of

quantity in MPU T1, frequency and quantity in MPU T2,

and purchasing price and variance of quantity in MPU

T3. Meanwhile, supplier 33320000 is managed in terms

of frequency and purchasing price in MPU T1, purchas-

ing price and its variance for one product and frequency

and quantity for the other product in MPU T2, purchasing

price and variance of quantity in MPU T3, and

purchasing price and its variance in MPU T4.

Purchasers should manage the selected suppliers to the

direction for improving their supply condition and for

reducing the supply risk.

Table 5

The change in supplier’s management directions over the period in time

MPU Supplier ID

37080000 33320000

Cluster Factors of management Cluster Factors of management

T1 4 Variance of quantity 7 Frequency (Zdelivery intervals) and purchasing price

T2 2 Frequency and quantity 4 Purchasing price and its variance

2 Frequency and quantity

T3 5 Purchasing price and variance of quantity 5 Purchasing price and variance of quantity

T4 – – 5 Purchasing price and its variance

G.H. Hong et al. / Expert Systems with Applications 28 (2005) 629–639638

5. Application and evaluation

The proposed model has been applied to supply selection

under the supply chain of the agriculture industry in Korea.

The agriculture products which farmers or the peasants’

association produce are supplied to purchasers such as

wholesalers and manufacturers. Then, the purchasers

process or package the products and delivery them to

customers. Because agriculture products are apt to decom-

pose and suppliers have different delivery intervals, harvest

quantities and levels of inventory facility, changes of supply

conditions according to the time period are larger than in

other industries. Moreover, customer needs change season-

ally over period in time. Therefore, if purchasers do not

consider these changes, they will not be supplied with

enough quantities on time from suppliers and they will have

to pay much money to supplement for the lack of quantities.

It is a very important issue how purchasers select their

suppliers under these conditions of the supply chain.

We design the method, which we have called the revised

Weber model, for comparing the performance of our model

as follows:

(1)

Criteria. Use the same criteria, that is, quality,

purchasing price and its variance, quantity and

Fig. 7. Comparing our method to the r

its variance and frequency (e.g. the number of

deliveries).

(2)

Pre-qualification. Use the SOM, clustering tool and

select the good clusters. Values of criteria used are the

average values during the total periods as other

researchers have not considered the period in time.

(3)

Final selection. Use the model of modifying the multi-

objective mathematical method of Weber (1998). That

is, we modified the multi-objectives to a single-

objective and the minimizing problem to a maximizing

problem for application of this case to the environment.

After analyzing the data of the past single year with our

model and the revised Weber method, we decided the

suppliers and their order quantities. Then we assigned the

order quantities to the selected suppliers for the next year and

compared the results of the two models in terms of the shortage

of order, revenue, and the number of managing suppliers. In

addition to the comparison of our model with revised Weber

model, we compared two cases in our method. One is that the

changing cost incurred when one supplier is changed to others

in any period is high and the other is that it is low.

First, our model manages a fewer number of suppliers

than the revised Weber method and the supply risk (shortage

evised Weber method.

G.H. Hong et al. / Expert Systems with Applications 28 (2005) 629–639 639

of order) of our model is lower than that of the revised

Weber method.

When both our method and the revised Weber method

select three suppliers as shown in the right of Fig. 7, the

suppliers which are selected from our model supplied the

order quantity on time without shortage of order. However,

many shortages of order occurred in the revised Weber

method in all MPUs (especially T2 in which the sales are

continuously high). The revised Weber method should

increase the number of suppliers to reduce the shortages of

order and it should pay a cost greater than that of our method.

Second, we can show that our model has more revenues

than the revised Weber model during all periods as shown in

the left of Fig. 7. The revised Weber method should increase

the number of suppliers for increasing the revenue.

However, the revenue of the revised Weber model may

not increase more than that of our model because the

maximum order quantities assigned to the best supplier are

limited by assigning order quantities to other suppliers and

the needs of the main customer are not considered.

Lastly, we can show the following facts when we compare

two cases of both high changing cost and low changing cost.

That is, the revenue obtained in the case of high changing cost

is less than the revenue obtained in the case of low changing

cost. However, purchaser maintains a good relation to

supplier during long-term in the case of high changing cost.

Therefore, we can show that the shortage of order obtained in

the case of high changing cost is less than the shortage of

order obtained in the case of low changing cost.

6. Conclusions

We suggested an effective supplier selection method for

maximizing revenue while satisfying the procurement

condition. We identified three problems and suggested a

method for their solution.

(1)

Evaluating the change in suppliers’ capability con-

ditions and customer needs (i.e. procurement condition)

over the period in time.

(2)

Defining the important criteria and using them step-by-

step to reduce the complexity of the problem.

(3)

Selecting suppliers to maximize revenue while satisfy-

ing the procurement condition and maintaining the

supplier-relationship for a long period in time.

The method consists of three steps: preparation, pre-

qualification, and final selection. In the preparation step,

we summarize the data by the unit period in terms of weeks

or packs of 10 days. In the pre-qualification step, we divide

the total analyzing period into several MPUs, identify

customer needs by MPU, and segment suppliers into several

groups sharing similar features. Then we evaluate the

features of each group. In the final selection step, we

finally select the optimal set of suppliers for maximizing

the revenue of the company and minimizing the supply risk

of the final selected set of suppliers.

We applied the method in the case of the agriculture

industry in Korea and compared it to the revised Weber

model in terms of the shortage of order, revenue, and the

number of managing suppliers. The shortage of order did

not occur in our method, but did in the revised Weber model

except the first MPU. Because of the shortage, the number

of managing suppliers was increased and the order amount

of each supplier was decreased in the revised Weber model.

As a result, the revenue of the revised Weber model was less

than that of our method. Our method informed us of the

changes of future supply condition and the directions for

managing the suppliers.

Although we applied the method in the case of the

Korean agriculture industry, we can further expand the

range of application to other industries in which the supply

condition undergoes excessive change over period in time.

Because many supplied products are aggregated to one final

product in several cases of these industries, it is very

difficult to measure the profit of the final product which is

affected by each supplied product.

We estimated the suppliers and their order quantities in

the next year by analyzing the data of the past year. However,

we could not determine whether it is better to use data of the

past single year for the analysis or data of the past several

years. Further research is necessary to answer this question.

References

Choy K. L., Lee W. B., & Lo V. (2002). Development of a case based

intelligent customer–supplier relationship management system, Expert

System with Application.

de Boer, L., Labro, E., & Morlacchi, P. (2001). A review of methods

supporting supplier selection. European Journal of Purchasing and

Supply Management, 7, 75–89.

Dickson, G. W. (1966). An analysis of vendor selection systems and

decisions. Journal of Purchasing, 2(1), 5–17.

Ha, S. H., & Park, S. C. (1998). Application of data mining tools to hotel

data mart on the intranet for database marketing. Expert System with

Applications, 15, 1–31.

Holt, G. D. (1998). Which contractor selection methodology? International

Journal of Project Management, 16(3), 153–164.

Kraljic, P. (1983). Purchasing must become supply management. Harvard

Business Review, 61(5), 109–117.

Lee, E.-K., Ha, S., & Kim, S.-K. (2001). Supplier selection and management

system considering relationships in supply chain management. IEEE

Transactions on Engineering Management, 48(3), 307–318.

Park, J. H., & Park, S. C. (2003). Agent-based merchandise management in

business-to-business electronic commerce. Decision Support System,

35, 311–333.

Talluri, S., & Sarkis, J. (2002). A model for performance monitoring of

suppliers. International Journal of Production Research, 40(16), 4257–

4269.

Weber, C. A., Current, J. R., & Desai, A. (1998). Non-cooperative

negotiation strategies for vendor selection. European Journal of

Operational Research, 108, 208–223.

Weber, C. A., & Desai, A. (1996). Determination of path to vendor market

efficiency using parallel coordinates representation: A negotiation tool

for buyers. European Journal of Operational Research, 90, 142–155.