Volume 3, Issue 10, October 2013 ISSN: 2277 128X

International Journal of Advanced Research in

Computer Science and Software Engineering
Research Paper
Available online at: www.ijarcsse.com
Life Insurance Recommender System Based on Association Rule
Mining and Dual Clustering Method for Solving Cold-Start Problem
Abdhesh Gupta Anwiti Jain
Department of Information Technology Department of Computer Science
Medicaps Institute of Science and Technology- Medicaps Institute of Science and Technology-
Indore, India Indore, India

Abstract— Recommender system based on web data mining is very useful, more accurate and provides worldwide
services to the user. Recommender systems are becoming very popular in recent years. In this paper a web
recommender system is proposed for life insurance sector based on web data mining using association rule which
supports the insurance needy as well as life insurance representative to select best suitable life insurance plan for any
particular person. Traditional recommender systems have being replaced by web mining techniques, but for new
profile customers these recommender systems are not suitable, this is known as cold-start problem. In this paper we
also proposed a solution for cold-start problem.

Keywords— Web data mining, Association rule mining, Apriori Algorithm, Cold-start, First-rater and Life insurance
recommendation system.

I. INTRODUCTION
Data mining can be defined as the process of selecting, exploring and modelling large amount of data to uncover
previously unknown patterns. In the insurance industry, data mining can help firms grow business advantage. For
example, by applying data mining techniques, companies can fully develop data about customers buying patterns and
behaviour- as well as gaining a greater accepting of their business to help minimize fraud, improve underwriting and
increase risk management. In this paper we explore the data mining technique for recommendation system using
association rule mining with some improvement in traditional recommendation system. In this paper we also discuss and
proposed solution for new customers that how to acquire new customers information into system and get best
recommendation for new customers. This is known as Cold-start problem [3].
Traditional recommendation methods can be classified into two main categories [4], Collaborative filtering and
content-based approach. Collaborative filtering techniques guess product preferences for a user based on the opinions of
other users. The opinions can be obtained openly from the users as a rating score or by using some inherent measures
from purchase records as timing records [5]. There are two approaches for collaborative filtering, user-based also known
as nearest-neighbours and item-based also know as model based algorithms. In the user-based method were the earliest
used [6]. They treat all user items by means of statistical techniques in order to find users with analogous preferences.
The advantage of these algorithms is the quick incorporation of the most modern information, but they have the
inconvenience that the search for neighbours in large databases is slow [7]. Item-based collaborative filtering algorithms
use data mining techniques in order to develop a model of user ratings, which is used to predict user preferences. In the
content based filtering method is based on content learn from the target items i.e. items are recommended by comparing
between their contents and user profile.
II. BACKGROUND
Association rule mining is one of the most important and well researched techniques of data mining, [1]. It aims to
extract interesting correlations, frequent patterns, associations or informal structures in the middle of sets of items in the
transaction databases or other data repositories. Association rules are broadly used in various areas such as
telecommunication, networking, market and risk management, inventory control etc. Association rule mining is to find
out association rule that assure the predefined least support and assurance from a given database. Association rule mining
is usually divided into two parts. One is to find those item sets whose occurrences exceed a predefined threshold in the
database, those item sets are called frequent or large item sets. The second part is to generate association rules from those
large item sets with the constraints of minimal confidence. Suppose one of the large item sets is Ln, Ln= {I1,I2,I3,….In},
association rules with this item sets are generated in the following way: the first rule is {{I1,I2,I3,….,In-1}}=>{In}, by
checking the confidence this rule can be determined as interesting or not. Then other rule are generated by deleting the
last items in the antecedent and inserting it to the consequent, further the confidences of the new rules are checked to
determine the interestingness of them. Those processes iterated until the antecedent becomes empty. The second part is
simple, so our main research is on first part.
III. ASSOCIATION RULE MINING
Insurance companies can use association rules in market analysis. Here the data analyses consist of information about
what policies customer purchases. The insurance company can generate association rules that show what different

© 2013, IJARCSSE All Rights Reserved Page | 951

 Gupta et al., International Journal of Advanced Research in Computer Science and Software Engineering 3(10),
October - 2013, pp. 951-954
policies are purchased with a specific policy. Based on these facts, company tries to capitalize on the association between
different policies with the same company is much more likely to renew than is a customer holding a single policy.
Similarly, a customer holding three policies is less likely to switch than a customer holding a single less than three. By
offering quantity discounts and selling bundled packages to customers, such as life security and investment policies, a
firm adds value and thereby increases customer loyalty, reducing the likelihood the customer will switch to a rival firm.
A database in which an association rule is to be found is viewed as a set of tuples, where each tuple contains a set of
items. Here there are some transactions and five items;
1. Life security,
2. Market based,
3. Tax benefit,
4. Investment and
5. Retirement plan.
We need to find the frequent product set with strong relation by Association Rule Mining and recommend these
products to web users to increase the cross-sales of life insurance sites.
According to Association Rule Mining, life insurance e-commerce can be displayed by following steps:-

A. Prepare data source for mining

In life insurance, there are basically three type’s data sources for mining: User information database, Transaction
database and Insurance plan database.
TABLE I: WEB DATABASE
User information Database Transaction Database Insurance plan Database
Field Name Type Field Name Type Field Name Type
UserId Char TransactionId Numeric PolicyId Char
UserName Char PolicyId Char PlanName Char
Gender Char UserId Char PlanDetials Char
DateOfBirth Date
Income Numeric
Contact Char

B. Apply Joins in Databases

To apply Association Rule Mining, we need to join database as shown in table
TABLE III: WEB DATABASE
UserId InterestedPlan TransactionId
User1 1,3,4 01
User2 1,2,4 02
User3 2,3,5 03
User4 1,2,4,5 04
User5 1,2,3 05

C. Find the frequent plan set and generate results

A program is set to automatically mine the frequent item set by Apriori algorithm [2].

Fig.1Apriori Algorithm
© 2013, IJARCSSE All Rights Reserved Page | 952
 Gupta et al., International Journal of Advanced Research in Computer Science and Software Engineering 3(10),
October - 2013, pp. 951-954
Algorithm I: Apriori Algorithm
The Apriori Algorithm is the most well known association rule algorithm and is used in most commercial products.
Input:
Li-1 //Large itemsets of size i-1
Output:
Ci //Candidates of size i
Algorithm:
Ci=Ø;
for each I € Li-1 do
for each J L≠ € i-1 do
if i-2 of the elements in I and J are equal then
Ck=Ck U{I U J};

1) Scan the reorganized database to find the support of items in table III and calculate the first-level frequent item
set.
TABLE III: FIRST LEVEL FREQUENT ITEM SET
Plan 1 2 3 4 5
Support 4 4 3 3 2
Frequent Plan Yes Yes Yes Yes Yes

2) Calculate second level frequent plan set from second- level candidate set as show in table IV

TABLE IV : SECOND-LEVEL FREQUENT PLAN SET

Plan Set {1,2} {1,3} {1,4} {1,5} {2,3} {2,4} {2,5} {3,4} {3,5} {4,5}
Support 3 2 3 1 2 2 2 1 1 1
Frequent Plan Yes Yes Yes No Yes Yes Yes No No No

3) Calculate third- level frequent plan set from third-level candidate set as show in table V

TABLE V : THIRD-LEVEL FREQUENT PLAN SET

Plan Set {1,2,3} {1,2,4} {1,2,5} {2,3,4} {2,3,5} {3,4,5}
Support 1 2 1 1 1 1
Frequent Plan No Yes No No No No

Result shows that {1, 2, 4} is the frequent plan set. If we keep calculating next level candidate plan set, a null set
appears, the data mining process finishes. Plan 1, 2 and 4 exist strong association rules.

IV. COLD-START PROBLEM AND PROPOSED SOLUTION

The first-rater or early-rater problem arises when it is impossible to offer recommendations about a policy that was
just incorporated in the system and, therefore, has few, or even none, evaluations from users. Analogously, such
drawback also occurs about him, it would be impossible to determine his behaviour in order to provide him
recommendations. Actually, this variant of the first rater problem is also referred as the cold-start problem [3].
An important business problem is the acquirement of new customers. In traditional approaches sales department make
stratagem and put their efforts to increase customer base, we can also improve results of this sales efforts by more
quantitative data mining approaches can lead to more focused and more successful results. A traditional sales approach
we target a group of persons and try to convince them for taking policies. The main problem with this approach is that
much of the marketing effort may not give good return. At the same point, sales become more difficult and greater
marketing budgets lead to lower and lower returns. Hence in that case it is important to identify population group among
already insured customers through which uninsured customers could be targeted. A statistical technique called “cluster
analysis,” sometimes used in the private sector to identify various market segments, was used to identify target groups of
uninsured persons based on the previous available data of policy holders. Clustering is a technique of partitioning or
segmentation the data into groups that might or might not be incoherent. The clustering usually accomplished by
determining the similarity among the data on predefined attributes. The most similar data are grouped into clusters. Since
clusters are not predefined, a domain expert is often required to interpret the meaning of the created clusters.
Insurance companies can create special catalogs targeted to various groups based on attributes such as income,
occupation and age as physical characteristic of possible customers. The company then can perform a clustering of
possible customers based on determined attribute values to create new catalogs. The results of the clustering exercise can
be then used by management to create special catalogs for different policies and distributed them to the correct target
population based on the cluster for that policy [2].
An insurance company can group its customers based on common features. Company management does not have any
predefined for this. Based on the outcome of the grouping they will target marketing and advertising campaigns to the
different groups for a particular type of policy.

© 2013, IJARCSSE All Rights Reserved Page | 953

 Gupta et al., International Journal of Advanced Research in Computer Science and Software Engineering 3(10),
October - 2013, pp. 951-954
TABLE VI : SAMPLE DATA
Age Occupation Income Education
35 Employee 20,000/- Graduate
25 Employee 10,000/- Graduate
55 Employee 30,000/- Post-Graduate
45 Employee 40,000/- Post-Graduate
40 Business 70,000/- Graduate
35 Business 80,000/- Graduate

For example (Refer table VI) we have customer following information Age, occupation, income and education. Here
we use dual clustering method. There are some limitations in single clustering. For example suppose advertising only for
policy of Life security, we could target the customers having less income and occupation as employee. Hence the first
group of people, is of younger employees having college degree, is suitable for Life security policies. The second group
has higher qualification and also higher income is suitable for tax benefit policies, while last group has businessmen with
higher income but low qualification and is suitable for investment policies [2]. As shown in table VI three clusters are
created according to their occupation and education. If new customer arrives to the recommendation system then we take
salary as a cluster parameter and according to salary we select average value of each cluster and put new customer in
nearest cluster, then we determine policy which is mostly preferred by cluster members and same policy will be
recommended to the new customer.
But if new customer salary is nearest to the more than one cluster average salary for example cluster 1 average salary
is 15,000/- and cluster 2 average salary is 35,000/- and new customer salary is 25,000/-, in that case it is difficult to
consider customer in a single cluster. To solve this problem we can use dual cluster method. In dual cluster method on
the basis of only one parameter (salary) if we cannot find single nearest cluster then we use second parameter for cluster
like in this example we can take age as a second parameter. By taking age as another parameter for clustering we can
decide final single cluster for new customer and recommend appropriate policy to the new customer.

V. CONCLUSION
In the insurance industry, web data mining can help firm gain business advantage mainly to support decision making.
The insurance companies need to know the essentials of decision making and web data mining techniques to compete in
the market of life insurance. In this work a web recommendation framework specially address to overcome critical
recommendation system problem. In this work some high level association rule mining method is used to retain existing
customer for new policy. Clustering method is used to attract and recommend policy to new customer (cold-start
problem). Dual clustering method is used to overcome the limitation of single clustering method which gives more
accurate and appropriate recommendation to solve cold-start problem.

REFERENCES
[1] Agrawal, R., Imielinski, T., Swami, A. Data Mining: A performance Perspective. IEEE Trans. Knowledge and Data
Engineering, vol, 5,6, 1993a, pp. 914-925.
[2] A.B. Devale, Dr. R. V. Kulkarni Applications of Data Mining Techniques in Life Insurance. International Journal of
Data Mining and Knowledge Management Process Vol.2, No.4 July 2012.
[3] Maria N. Moreno, Saddys Segrera, Vivian F Lopez, Maria Dolores Munoz and Angel Luis Sanchez, Mining
Semantic Data for Solving First-rater and Cold-start Problems in Recommender system ACM IDEAS 11 2011,
September 21-23.
[4] Lee, CH., Kim, Y.H., Rhee, P.K. 2001. Web personalization expert with combining collaborative filtering and
association rule Mining Technique. Expert System with Applications 21. 131-137.
[5] Sarwar, B., Karypis, G., Konstan, J., Riedl, J. 2001. Item-based Collaborative Filtering Recommendation Algorithm.
Proceedings of the tenth International World Wide Web Conference, 285-295.
[6] Resnick, P., Lacovou, N., Suchack, M., Bergstrom, P. and Riedi, J. 1994. Grouplens: An open architecture for
collaborative filtering of netnews. Proc of ACM Conference on Computer Supported Cooperative Work, 175-186.
[7] Schafer, J.B., Konstant, J.A. and Riedl, J. 2001. E-Commerce Recommendation Applications. Data Mining and
knowledge Discovery, 5, 115-153.

© 2013, IJARCSSE All Rights Reserved Page | 954