LABELING CUSTOMERS USING DISCOVERED KNOWLEDGE CASE STUDY: AUTOMOBILE INSURANCE INDUSTRY

International Journal of Managing Value and Supply Chains (IJMVSC) Vol. 3, No. 3, September 2012
DOI: 10.5121/ijmvsc.2012.3302 13
LABELING CUSTOMERS USING
DISCOVERED KNOWLEDGE
CASE STUDY: AUTOMOBILE INSURANCE
INDUSTRY
SARAH AZADMANESH 1
and MOHAMMAD J. TAROKH2
1
MSc student in industrial department, K.N.Toosi University,Tehran, Iran
saraazadmanesh@gmail.com
2 Associated professors in industrial department, K.N.Toosi University,
Tehran, Iran
mjtarokh@kntu.ac.ir
1. Abstract
In this paper, we used the knowledge discovery in databases and data mining, one of the data-based
decision support techniques to help labeling customers in the automobile insurance industry. In most data
mining application cases, major tasks including data preparation, data preprocessing, data
transformation, data mining, interpretation, application and evaluation, are required. The results of a
case study are presented that knowledge discovery of databases and data mining is used to explore
decision rules for an automobile insurance company. The decision rules can be used to label the
customers as “bad” or “good” for insurance policies.
2. Keywords
labeling customers, discovered knowledge, decision tree, ID3
3. Introduction
Customer relationship management (CRM) includes a set of processes and it enables systems to
support a business strategy to build long term, profitable relationships with specific customers
[11]. Customer data and information technology (IT) tools form the foundation of any successful
CRM strategy. Also, the rapid growth of the Internet and its connected technologies has
extremely increased the opportunities for marketing and transformed the way that relationships
between companies and their customers are managed [14].
From the architecture point of view, the CRM framework can be classified into operational and
analytical ([1]; [8]; [21]). Operational CRM apply to the automation of business processes, while
analytical CRM refers to the analysis of customer characteristics and behaviors to support the
organization’s customer management strategies. Analytical CRM could help an organization to
better separate and more effectively allocate resources to the most profitable group of customers.
Data mining tools are a popular means of analyzing customer data within the analytical CRM
framework. Many organizations have collected and stored a wealth of data about their current
customers, potential customers, suppliers and business partners. However, the inability to
explore valuable information hidden in the data prevents the organizations from transforming
these data into valuable and useful knowledge [1]. In order to effectively reveal information in
handling data, information technology has introduced the development and application of the
techniques of knowledge discovery (KD) and data mining (DM). KD/DM is basically a data-
oriented method that merges database management, knowledge representation, and machine

14
learning [3]. It is introduced with both descriptive and predictive ability to discover patterns in
historical data that was unknown before, but meaningful and decision-supportable. In response
to its applications in business, KD/DM has been widely employed in support of management
decisions [6]. Pitta (1998) highlighted KD/DM as an important tool that marketers can use to
explore patterns in databases, whereas also emphasizing the one-to-one marketing strategy [16].
Feelders, Daniels, and Holsheimer (2000) presented a fundamental concept for data mining with
aiming at its application process [5]. More importantly, the applications in different areas of
business shown in literature in the past few years have also witnessed the increased use of
KD/DM. With extensive customer data, data mining techniques can provide business
intelligence to create new opportunities [14].
Technically, there is much to argue with in order to successfully implement a KD/DM
application plan. It is realized that when KD/DM is applied in a field, it systematically pursues
six phases: identifying the problem, collecting datasets, preprocessing collected datasets, mining
preprocessed datasets, interpreting and using discovered informatics, and also evaluating the
discovered knowledge. For identifying the problem, one should specifies the critical question(s)
and/or the issues that are associated with the improvement of decision-making capability, like
‘who are the potential customers?’
Using data mining tools in CRM is a novel trend in the global economy. A competitive CRM
strategy is based on analyzing and understanding customer behaviors and characteristics, for
acquiring and retaining potential customers and maximizing customer value. Proper data mining
tools, which are good at exploring and identifying practical information and knowledge from
customer databases, are one of the best supporting CRM decisions tools [1].
Iranian companies due to the exclusion of services in the hand of governmental section aren’t
familiar with the importance of customer data and targeting the best customers. They just store
their customer data without any expectation and concern. Case company is one of the largest
formal governmental companies and the management is looking forward to any tool that helps
improving their competitive advantage and keeping their existing position. Insurance companies
have limited financial resources and they can’t pay all of the recompenses so they should know
their customer and their accompanied risk. Labeling customer help them to target the right
customer with the lower risk and less behavior anomaly. Labeling customers who bought the car
insurance policy can be helpful in advertising the third-party insurance policy too, it’s
mandatory in Iran. They can use the result in targeting the “good” customer for other insurance
policies. We are going to label customers with the help of data mining.
This paper is organized as follows: Section 2 describes the literature review where KD/DM is
utilized to explore knowledge in multiple issues and in insurance industry. Discussions are
presented in Section 3. Section 4 concludes this research and addresses future research.
4. Literature review
In this section the literature is viewed from two perspectives. One, the published papers that used
ID3 decision tree in their applications and the other is the published papers about insurance
industry that used data mining techniques and was published from 2000 to 2011.
4.1. Papers that used ID3 decision tree
Prasad et al ([15]) used Machine learning algorithms such as Auto-associative memory neural
networks, Bayesian networks, ID3 and C4.5 for diagnosing the Asthma .They presented a
comparative study among these algorithms with the use medical expert systems on patient data.
They gathered the clinical signs and symptoms of asthma of patients from various resources.
Based on the analysis of such data, it was found that the Auto-associative memory neural

15
networks are one of the best among the remaining algorithms in terms of efficiency and accuracy
of identifying the proper disease.
Zhilin & Wang ([30]) proposed a fault diagnosis approach based on data mining technologies, in
order to improve the maintenance quality and efficiency. By making full use of data stream, they
firstly extract fault symptom vectors by processing data stream, and then establish a diagnosis
decision tree through the ID3 decision tree algorithm, and finally stored the link rules between
faults and the related symptoms into historical fault database as a foundation for the fault
diagnosis. This database provides the basis of trend judgments for a future fault.
Mohanty et al ([12]) employed ID3 decision tree (J48) to predict the quality of a web service
based on a set of quality attributes. Their experiments were carried out on the QWS dataset.
They applied 10-fold cross-validation to test the efficacy of the models. The J48 and TreeNet
techniques outperformed all other techniques by yielding an average accuracy of 99.72%. They
also performed feature selection and found that web-services relevance function (WSRF) is the
most significant attribute in determining the quality of a web service. Later, they performed
feature selection without WSRF and found that Reliability, Throughput, Successability,
Documentation and Response Time are the most important attributes in that order. Moreover, the
set of ‘if–then’ rules yielded by J48 and CART can be used as an expert system for web-services
classification.
In Ture et al ([22]) , they analyzed the simultaneous relationship between risk factors for breast
cancer with five decision tree algorithms(classification and regression tree (C&RT), Chi-squared
automatic interaction detector (CHAID), quick, unbiased, eﬃcient statistical tree (QUEST),
Commercial version 4.5(C4.5) and Interactive Dichotomizer version 3 (ID3)). They compared
the relative impacts of each risk factor for breast cancer in the multivariate analysis models.
Ture et al ([23]) determined new diagnostic indexes for the distinguishing of subgroups of breast
cancer patients by the decision tree algorithms techniques (C&RT, CHAID, QUEST, ID3,
C4.5and C5.0) and analyzing breast cancer patients with Cox regression for disease-free survival
(DFS). A retroactive analysis was performed in 381 diagnosed patients with breast cancer. Based
on these diagnostic factors, new diagnostic indexes for C&RT, CHAID, QUEST, ID3, C4.5 and
C5.0 and Cox regression were obtained. Prognostic indexes showed a good degree of
classification, which shows improvement, is possible using standard risk factors. They acquire
that C4.5 has a better performance than C&RT, CHAID, QUEST, ID3, C5.0 and Cox regression
for determining risk groups using Random Survival Forests (RSF).
Emekci et al ([2]) proposed a new privacy preserving distributed decision tree learning algorithm
based on the ID3 algorithm [17] and Shamir’s secret sharing [20]. Their suggested algorithm is
scalable from computation and communication cost point of view, and hence it can be run even
with large number of parties involved. Their goal was implementing constructing decision trees
over an arbitrary number of distributed data sources in a privacy preserving manner based on the
ID3 algorithm.
4.2. Using DM in insurance industry
Wang et al ([25]) combined quantitative and qualitative analysis, and created a multi-criteria
comprehensive evaluation model based on fuzzy mathematic. They proposed it for calculation of
risk coefficient for adjusting basic premium rate of Construction Work Safety Liability
Insurance (CWSLI).
Lin ([10]) analyzed consumer switching behaviors from consumer psychology and interactions
with the society point of view. His results exhibited that when service failures are more severe,
higher emotional intelligence consumers exhibit lower intent to switch than those with lower
emotional intelligence. Similarly, the more severe the service failures, the more the internal

16
locus-of-control orientation consumers exhibit intention to change their existing utilization than
those with external locus-of-control orientation.
Donkers et al ([3]) studied the abilities of a range of models for CLV prediction in the insurance
industry. The simplest models can be building at the customer relationship level, i.e. aggregating
all services. The more complex models concentrated on the individual services, noticing
explicitly to cross buying, but also retention. They discovered that the simple models perform
well. Richness of relationship development is expected from more complex models.
Unexpectedly, they realized that more complex models do not lead to substantially better CLV
predictions.
In [28], Wu et al. used the knowledge discovery in databases and data mining (KDD/DM) to
help targeting customers in the insurance industry. A case study of KDD/DM was performed to
explore decision rules from company’s data base. They suggest that decision rules could be used
for investigating the potential customers for an insurance product (existing or new).
Roh at al ([18]) investigated the CRM implementation success factors from three perspectives:
efficiency in IS, customer satisfaction in marketing, and firms aggregated profitability. The result
of their study discovered multi dimensional metrics for factors that affect profitability through
CRM that are reliable. Their measurement model analysis indicated that their suggested metrics
have a relatively high degree of validity and reliability.
In ([24]), Verhoef and Donkers introduced a predicting potential value for a current customer
model. Their study was the first to focus on the modeling and predicting the potential value of
multi-service provider customers. Second, they provide a framework for CRM-managers in
multi-service industries, which could be used for predicting customer potential value. This
framework considered the data limitations that a company usually has, with using socio-
demographic information and transaction information from the customer database only.
5. Application framework
The application framework that we used in this paper is illustrated in Figure 1.It consists of two
major components. The first component deals mainly with the task of knowledge discovery and
feedback from the application case. The second one is the discussions drawn from the
application case, as a reference for any case that is the same as or similar to the current study.
The application features are described in Table 1, including the objective of the application, the
size of database used, the way to prepare data (e.g. attribute selection) that was mined, the way
to granulate attributes, the mining mechanism used the representation of discovered knowledge,
and the purpose of the discovered knowledge. Details are described below.

17
Figure 1. Application framework
Table 1. Features of the company case
Features Description
Objective Labeling customers in case company (automobile
insurance industry )
Size of database used 19,336
Data preparation 1. Transformation from Persian to English
2. 2 Numeric and 3 text attributes and 1
semantic class selected by a discussion of the case
company
3. 4 Concepts in class attribute
Data transition Listed in Table 2 by a discussion of
case company
Mining mechanism used Induction-based ID3
Representation of discovered Decision rules via classification
Knowledge Labeling customers for insurance
products
5.1. Case company data and data preparation and selection
The data that were used in this study were selected from a former governmental insurance
company in Iran (now a private company). The original data from the case company was in
Persian format, so we first transformed these data from Persian to English. The total number of
attributes was fifty. It is known that too many attributes involved in data mining will harden the
interpretation or cause meaningless discovered information. Therefore, by in-depth discussion
with domain experts, we eliminated some of the attributes and finally came to a conclusion of 6
attributes, namely (1) Insured sex (sex of the person who bought an insurance policy) (2) History
kind (is there a history about the car or its owner in the data base) (3) Value group (estimation of
car value by insurance company granulating in groups like Table 2) (4)Police report (when
police came to accident scene how counterparts act, agreed on the damage and the amount of
recompense or not agreed, or they didn’t call the police or other conditions) (5)Group of
compensation (the amount of money that insurance company paid to insurer after accidents for
compensation, granulated in groups like Table 2), and finally Customer labels, type of customer.
The attribute of class in the database showed 2 types of Customer Label (see Table 2).
Furthermore, because some attributes contained numerical value, Data transition was necessary

18
for the mining mechanism performing the knowledge explosion operation [29]. Generally, the
number of granules for an attribute will greatly affect the information granularity definition and
finally discovered decision rules [27]. However, a dilemma occurs with this decision since it is
very possible that the generated decision trees will be too complex to interpret if the numbers of
granules are too large. Specifically, the decision trees that are generated using induction-based
technique will be too large to interpret their patterns if there are too many group of granulation
for labeling. If too small, a group of granulation may not be able to contain acceptable number of
instances, and consequently the discovered results may be meaningless. In spite of full freedom
for a granulation method, the number of granules should be carefully chosen to eliminate
unnecessary problems [28]. By a discussion with the case company experts, the granulation
operation was based on the granulation regulations that are listed in Table 2. For example, the
attribute of Value group was granulated into twenty nine levels, from value A, B,C,D to ZC. The
defined ranges were 0 to 38500000 Rials(Iran currency) for the value A , 38500000 to 49000000
for value B, 49000000 to 59000000for value C, 59000000to 63500000 for value D, and at last
from 1000000000 and more for value ZC, respectively. Note that the class attribute contains 2
types of customer label that are denoted good and bad.
Table 2. Data transition regulation for attributes
Attribute
number
Names Data transition regulation
1 Insured sex female male
2 History kind New damaged No history of
damage
3 Value group value A Value B value C value D value E value ZC
38500000 49000000 59000000 63500000 69000000 … 1950000000
4 Police report Not exist agreed Not agreed other
5 Group of
compensation
Group one Group two Group three Group four Group thirty
632000 737709 832500 895000 … 75500000
6 Customer
Label
Good Bad
5.2. Data mining
3.2.1 Mining mechanism used
In this paper, we used the classification method as the mining mechanism. ID3, an important and
most widely applied classification method, was used to mine decision rules in the collected
database. The ID3 algorithm, introduced by [17], has been wildly used to help measure the
information entropy for a set of data under the consideration of multiple classes ([13]; [19];
[26]). ID3, C4.5, and CART adopt a greedy (i.e., non back tracking) approach through the given
sets to test each attribute at every tree node and decision trees are constructed in a top-down
recursive divide-and-conquer manner. In order to select the attribute that is most useful for
classifying a given sets, ID3 uses information gain measure. This measure is based on
pioneering work by Claude Shannon on information theory, which studied the value or
“information content” of messages. Let Node N represent or hold the tuples of partition D. The
attribute with the highest information gain is chosen as the splitting attribute for node N. This
attribute minimizes the information needed to classify the tuples in the resulting partitions and
reflects the least randomness or “impurity” in these partitions. Such an approach minimizes the
expected number of tests needed to classify a given tuple and guarantees that a simple (but not
necessarily the simplest) tree is found. The expected information needed to classify a tuple in D
is given by

19
(1)
where is the probability that an arbitrary tuple in D belongs to class and is estimated by
| |/|D|. A log function to the base 2 is used, because the information is encoded in bits. Info(D)
is just the average amount of information needed to identify the class label of a tuple in D. Note
that, at this point, the information we have is based solely on the proportions of tuples of each
class. Info(D) is also known as the entropy of D.
Now, suppose we were to partition the tuples in D on some attribute A having v distinct values,
{ , , … , }, as observed from the training data. If A is discrete-valued, these values
correspond directly to the v outcomes of a test on A. Attribute A can be used to split D into v
partitions or subsets, { , , … , },where contains those tuples in D that have outcome
of A. These partitions would correspond to the branches grown from node N. Ideally, we
would like this partitioning to produce an exact classification of the tuples. That is, we would
like for each partition to be pure. However, it is quite likely that the partitions will be impure
(e.g., where a partition may contain a collection of tuples from different classes rather than from
a single class). The amount of information is measured by
(2)
The term acts as the weight of the jth partition. InfoA(D) is the expected information required to
classify a tuple from D based on the partitioning by A. The smaller the expected
information (still) required, the greater the purity of the partitions. Information gain is defined as
the difference between the original information requirement (i.e., based on just the proportion of
classes) and the new requirement (i.e., obtained after partitioning on A). That is,
Gain(A) = Info(D) - (D) (3)
In other words, Gain(A) tells us how much would be gained by branching on A. It is the
expected reduction in the information requirement caused by knowing the value of A. The
attribute A with the highest information gain, (Gain(A)), is chosen as the splitting attribute at
node N. This is equivalent to saying that we want to partition on the attribute A that would do
the “best classification,” so that the amount of information still required to finish classifying the
tuples is minimal (i.e., minimum (D)).[7]
All instances in a granulated database have to consider because they all have to contribute to the
generation of decision rules. The discovery process begins partitioning via the value that the
attribute work to form a decision tree, and therefore the decision rules can be generated. This
process is repeated until same result is obtained to return decision rules.
3.2.2. Learning accuracy test
Table 3. Result accuracy
Class TP Rate FP Rate Precision F-Measure ROC Area
Good 0.969 0.035 0.936 0.952 0.997
Bad 0.965 0.031 0.984 0.974 0.997
Weighted Avg. 0.967 0.032 0.967 0.967 0.997

20
Table 4. Confusion matrix
A B Classified as
6451 204 A=Good
441 12240 B= Bad
ID3 has shown its prediction accuracy and suitability but we test our instances for accuracy too.
As shown in table 3 for class good, True Positive (TP) is 0.969 and False Positive (FP) is 0.035
and F-Measure is 0.952 and precision is 0.936. For Bad class the TP and FP are 0.965, 0.031,
respectively and F-Measure is 0.974 and the precision is 0.984. The weighted average precision
is 0.967 it’s a very good result and its very acceptable. From all 19,336 instances, 18691
(96.6643 %) instances were correctly classified and 645 (3.3357 %) instances were incorrectly
classified.
Table 5. Result characteristics
Correctly Classified Instances 18691 96.6643 %
Incorrectly Classified Instances 645 3.3357 %
Mean absolute error 0.0445
Root mean squared error 0.1491
Relative absolute error 9.8535 %
Root relative squared error 31.3905 %
Total Number of Instances 19336
3.2.3. Discovered knowledge and interpretation
The study reached the knowledge discovery for the all granulated database cases (19,336 cases).
After executing knowledge discovery and data mining application, 3780 rules was discovered. In
figure 2 sample of discovered knowledge (tree) with one level of tree (if it is necessary) is
shown. The number of n implicitly indicates the number of cases in a rule. More precisely, the
more the number of n, a decision rule has, the more reliable it is because it covers more cases in
the transaction database. We think that management will prefer a decision rule if it is supported
by a higher number of cases. Since there were 19,366 cases used and 3780 rules discovered,
each rule set was supported by 88.8% of accuracy in average. Part of rule sets is illustrated in
Fig. 3. These decision rules can be used to serve as an advisor to help in labeling customer based
on his/her personnel characteristics.
Figure 2. Part of discovered rule set
Noteworthy observations based on the customers’ characteristics were also described below.
I. The default label for customers is bad. It is not odd since most of customers were
labeled bad before performing the data mining application.

21
II. The Group of compensation attribute of cases was the most important attribute for
classifying customers. In the first place, based on this attribute company can classify
its customer. Granulation in this attribute is a very important factor, with no suitable
granulation the accuracy of model can be decline dramatically. In this study we first
used a C5.0 decision tree for discovering the break point in this attribute and value
group (the value of the automobile) and then based on this break point we derived the
proper granulation measures. It’s obvious that when the amount of compensation
increased more customers labeled as bad.
III. After the Group of compensation attribute the value group attribute is the second
important attribute for classifying. For this attribute, like the Group of compensation
attribute, we used C5.0 decision tree. Suitable granulation of this two attribute has a
great effect on the accuracy of model. With increase in value of automobile
customers label is change from bad to good, it is sensible that with valuable
automobile people drive more carefully.
IV. History and sex attribute of a customer has the less importance in our model. These
two attribute have a 3 and 2 possible value, respectively. Most of Men labeled as bad
and women have almost equally labeled as bad and good. For history attribute it’s
like that the default label for all values is bad and just in ‘new’ value less than half of
them labeled as good. This two attribute may have not a significant effect on labeling
customer but they are important in classifying them.
V. Police report is the least important attribute for classifying a customer. If the police
report wasn’t exist, customer labeled good equally as bad, and if both side not agreed
with each other most of them labeled as bad, it’s not shocking because they couldn’t
agree on the amount of recompense because in most of these cases the damage is
extended.
As shown in figure4, when a car is new and value group is C (third group – one of the lowest
group of values) and the amount of recompense is the lowest one the customers mainly labeled
as “good”, although it’s not surprising cause two main attribute for labeling customers is the
value group and the group of compensation and in this rule both of them are from the lowest
Figure 3. Part of discovered knowledge

22
groups. For other history kinds the label is good too, due the previous discussed logic.
For middle groups of value like G, H when group of compensation is around the middle, they are
labeled as good too. As shows in the above rules, one can understand that when the amount of
both of group of compensation and value group are near each other, customers are labeled as
“good”, and when the value group is higher than the group of compensation the customers
labeled as “good” too. Finally when the value group is low and the group of compensation is
high the customer label is “bad”.
The importance of labeling customers is for the future when they come for buying another
insurance policy for their car, the insurance company has a back ground and can predict the
customer label based on his/her demographics and with that in mind can decide whether
reinsured the car or not. The other benefit of predicting a customer label is for third-party
insurance that it is mandatory in Iran and every vehicle must have one. When a customer is
predicted as bad based on his/her car insurance it means the recompense amount was or would
be higher than what he/she pays or going to pay to the insurance company, that means she/he
had or can have accident or accidents that causes a lot of damage and those accident were
serious and can be fatal, so he/she have a high risk and it would be wise to not insure her/his car
or insuring the car with a higher tariff.
6. Discussion
As shown in Fig. 1, in this discussions we mainly focus on three things: (1) the experiences of
Knowledge Discovery and Data Mining (KD/DM) application in labeling customers for the
automobile insurance industry,(2) the usability of KD/DM in supporting decision-making, (3) a
lesson from theory to practice. First, the KD/DM technique is basically a data-based pattern used
as a management tool in support of decision-making. Although the experience-oriented decision-
making has been utilized for a long time, and will continue to have an important role in
supporting the management, the decision support mode has incrementally shifted from
experience-oriented to information-oriented. At the beginning of the first interview, we
experience that the case company seemed unfamiliar with what KD/DM is about and what
objective KD/DM can get. In regard to the usability of KD/DM, we have presented the
discovered knowledge such as what a rule say, how to explain a rule, what information a rule
can get, etc. The usability of the discovered knowledge presented in our research showed that at
this stage of KD/DM could be a possible path for improvement of labeling customers. For the
measuring the effect of the discovered knowledge at the second stage, it will take time for the
case company to collect data and do comparison. For example, does the discovered knowledge
significantly help predict the customer label for the next year (or next 2, 3, 4,., years)? The
KD/DM techniques, application process and insurance knowledge are all the effective factors
that need to be considered carefully. More importantly, the main purpose of the KD/DM
technology is not the technology itself, but to help improve management quality, in particular
quality of decision-making. This needs continuous involvement of domain researchers and
experts with practitioners. An insurance company must be aware of the ability of learning in its
data, its potential benefits.
In Iran most of economical sections were exclusively under governmental jurisdiction and now
the government gives its jurisdiction to the private sections for economical release. The case
company was one of the largest governmental insurance companies in Iran. After this exclusion
break down the case company needed new techniques for staying at the top. Because of the
governmental exclusion, Iranian companies never had an experience of how they can use their
customer data for attracting new ones and retaining the old ones. After the exclusion break down
they are eager to use these data for making profit.

23
7. Conclusion
In this paper, we have described the major tasks necessitated for the KD/DM application, the
case application was in an automobile insurance company. Three points obtained from this
application case are addressed: (1) KD/DM application experiencing in labeling customers for
the automobile insurance industry; (2) the usability of KD/DM techniques in supporting
insurance decision-making; (3) a lesson from theory to practice. With fast growth in the volume
of information systems collected data, more and more marketers are using data-based decision
support tools to enhance the efficiency and effectiveness of their marketing decisions. Using
KD/DM technology may need domain knowledge and KD/DM professionals to succeed because
it is a highly domain-specific task. Data preparation and transformation is the most time-
consuming phase of KD/DM application. From the application viewpoint, the most important
part may be the methodology that can both deal with data preparation, transformation and carry
out interpretation, utilization and evaluation of knowledge on a domain-by-domain basis.
8. Acknowledgement
The authors thank the editor and anonymous reviewers. The authors would also like to thank the
participating insurance company employees for their cooperation.
References
[1] Berson, A, Smith, S, & Thearling, K. (2000). Building data mining applications for CRM. McGraw-Hill.
[2] Emekci F , Sahin, O, Agrawal, D, El Abbadi , A.(2007).Privacy preserving decision tree learning over multiple
parties. Data & Knowledge Engineering 63 (2007) 348–361
[3] Donkers , B, Verhoef, P, Jong, M. (2007) . Modeling CLV: A test of competing models in the insurance
industry. Quant Market Econ (2007) 5:163–190
[4] Fayyad, U, Stolorz, P. (1997). Data mining and KDD: Promise and challenge. Future Generation Computer
Systems, 13(2–3), 99–115.
[5] Feelders, A, Daniels, M, Holsheimer, M. (2000). Methodological and practical aspects of data mining.
Information & Management, 37, 271–281.
[6] Han, J, Fu, Y. (1999). Mining multiple-level association rules in large databases. IEEE Transactions on
Knowledge and Data Engineering, 11(5), 798–805.
[7] Han, J ,Kamber, M .(2006) . Data Mining: Concepts and Techniques. Elsevier.
[8] He, Z., Xu, X., Huang, J, Deng, S. (2004). Mining class outliers: Concepts, algorithms and applications in CRM.
Expert Systems with Applications, 27, 681–697.
[9] Hirota, K, Pedrycz, W. (1999). Fuzzy computing for data mining. Proceedings of the IEEE, 87(9), 1575–1600.
[10] Lin, W.( 2009).Service failure and consumer switching behaviors: Evidence from the insurance industry. Expert
Systems with Applications 37 (2010) 3209–3218.
[11] Ling, R., & Yen, D. C. (2001). Customer relationship management: An analysis framework and implementation
strategies. Journal of Computer Information Systems, 41, 82–97.
[12] Mohanty R, Ravi V, Patra M. (2010) Web-services classification using intelligent techniques . Expert Systems
with Applications 37 5484–5490
[13] Murthy, S. (1998). Automatic construction on decision three from data: A multi-disciplinary survey. Data
Mining and Knowledge Discovery, 2, 345–389.
[14] Ngai E., (2005) .Customer relationship management research (1992-2002): An academic literature review and
classification. Marketing Intelligence & Planning, Vol. 23 Iss: 6, pp.582 – 605.
[15] Prasad, B. D. C. N., Krishna, P. E. S. N., Sagar,Y.(2011). A Comparative Study of Machine Learning
Algorithms as Expert Systems in Medical Diagnosis (Asthma).Communications in Computer and Information
Science, 131,570-576.
[16] Pitta, D. (1998). Marketing on-to-one and its dependence on knowledge discovery in databases. Journal of
Consumer Marketing, 15(5), 468–480.
[17] Quinlan, J. (1986). Induction of decision tree. Machine Learning, 1, 81–106.
[18] Roh ,T , Ahn, C, Han I. (2005).The priority factor model for customer relationship management system success.
Expert Systems with Applications 28 (2005) 641–654.
[19] Sestito, S, Dillon, T. (1994). Automated knowledge acquisition. New York: Prentice Hall.
[20] Shamir , A .(1979) . How to share a secret, Communications of the ACM 22 (11) 612–613.

24
[21] Teo, T, Devadoss, P, Pan, S. (2006). Towards a holistic perspective of customer relationship management
implementation: A case study of the housing and development board, Singapore. Decision Support Systems, 42,
1613–1627.
[22] Ture, M, Tokatli F, Kurt I(2008 a).Using Kaplan–Meier analysis together with decision tree methods (C&RT,
CHAID, QUEST, C4.5 and ID3) in determining recurrence-free survival of breast cancer patients. Expert
Systems with Applications 36 2017–2026.
[23] Ture, M , Tokatli F , Omurlu I(2008 b) .The comparisons of prognostic indexes using data mining techniques
and Cox regression analysis in the breast cancer data. Expert Systems with Applications 36 8247–8254.
[24] Verhoef, P, Donkers, B. (2001). Predicting customer potential value an application in the insurance industry.
Decision Support Systems, 32, 189–199.
[25] Wang,H.,Ye,G.,Ju,C.(2011). Risk based determination of the prenium rate construction of work safety liability
insurance .Computational Risk Management,2, 113-120.
[26] Wu, C.(2003a). Data mining applied to material acquisition budget allocation for libraries: Design and
development. Expert Systems with Applications, 25(3), 401–411.
[27] Wu, C. (2003b). On the granulation simplicity for the decision rule discovery in databases: EWI vs. EFI.
International Journal of Science and Technology, 14, 28–36.
[28] Wu, C, Kao, S, Su, Y, Wu, C. (2005). Targeting customers via discovery knowledge for the insurance industry.
Expert Systems with Applications, 29, 291–299.
[29] Wu, X, Urpani, D. (1999). Induction by attribute elimination. IEEE Transactions on Knowledge and Data
Engineering, 11, 805–812.
[30] Zhilin, Z., Wang, P.(2010). Fault Diagnosis of Automobile ECUs with Data Mining Technologies. Applied
Mechanics and Materials ,40-41, 156-161.

LABELING CUSTOMERS USING DISCOVERED KNOWLEDGE CASE STUDY: AUTOMOBILE INSURANCE INDUSTRY

More Related Content

What's hot (19)

Similar to LABELING CUSTOMERS USING DISCOVERED KNOWLEDGE CASE STUDY: AUTOMOBILE INSURANCE INDUSTRY (20)

Recently uploaded (20)

LABELING CUSTOMERS USING DISCOVERED KNOWLEDGE CASE STUDY: AUTOMOBILE INSURANCE INDUSTRY