TWO PARTY HIERARICHAL CLUSTERING OVER HORIZONTALLY PARTITIONED DATA SET

International Journal of Data Mining & Knowledge Management Process (IJDKP) Vol.7, No.3, May 2017
DOI: 10.5121/ijdkp.2017.7303 33
TWO PARTY HIERARICHAL CLUSTERING OVER
HORIZONTALLY PARTITIONED DATA SET
Priya Kumari1
and Seema Maitrey2
1
M.Tech (CSE) Student KIET Group of Institution, Ghaziabad, U.P ,
2
Assistant Professor KIET Group of Institution, Ghaziabad, U.P
ABSTRACT
Data mining is a task in which data is extracted from the large database to make itin an understandable
form or structure so that it can be used for further use. In this paper we present an approach by which the
concept of hierarchal clustering applied over the horizontally partitioned data set. We also explain the
desired algorithm like hierarichal clustering, algorithms for finding the minimum closest cluster. In this
paper wealso explain the two party computations. Privacy of any data is the most important thing in these
days hence we present an approach by which we can apply privacy preservation over the two party which
are distributing their data horizontally. We also explain about the hierarichal clustering which we are
going to apply in our present method.
KEYWORD
Two party computations, Partitioning, clustering, k-means algorithm, Hierarichal clustering.
1. INTRODUCTION
Data mining is a very current research area in these days only because of its ability to extract the
data from a large data set very efficiently. Data mining is a field in which the main aim is to
extract or mine knowledge from a large amount of data [1]. In data mining generally the
processing is done over the large volume of data that is stored in a database and search for pattern
and relationships inside the data. Privacy is also the main point of focus in these days in between
the researchers all of us have some data that we don’t want to share with anyone hence whenever
the situation arises that we wants to secure our data from others then we use have some
approaches like association rule mining, classification and clustering. In this paper we are going
to use clustering approach.
Clustering of data is a method by which the similar kind of cluster are grouped together and one
by one each attribute comes to any cluster in the end of the approach. When we are dealing with
some sensitive information then the privacy issue is a major concern because if any of the
information is leak or compromise then that may result to effect or harm to individual or financial
losses to any well stablished organisation. Clustering is used widely in many real time areas like
in financial affairs, marketing, medical, chemistry, insurance, machine cleaning, data mining etc
[2,3].

34
2. PRELIMINARIES
In this section we are going to present some preliminaries. We first introduce the partitioning
approach by which we partition the data set. Then we explain about how to party works and after
then the main approach or algorithms which we are going to implement to fulfil our approach.
2.1 Two party computation
Two party computationsare an approach in which mainly two party involve in computation.
These two parties have their own data set in an equal amount but they don’t expose their data to
its corresponding party. In this way they a form an distance matrix and share their distance matrix
to each other not the original data set by merging the two distance matrix of each party we come
to a single matrix by which we can easily get the solution of our queries because these distance
matrices compute the smallest distance of each cluster with the help of cluster center so any user
get their result in a very reasonable time [4]. Two party computation query model which consist
of mainly four entities these are following:-
1. Randomizer.
2. Computing engine.
3. Query front end engine.
4. Individual Database.
In this the randomizer and computing engine are comes in primary engine. The query front end
engine which is responsible for receiving all of the queries from the users and then it forward
these queries to the randomizer. An encoded query which is normally contains the type of query
to its computing engine. The computing engine coordinates that query to individual database for
computing the result of query.In our approach we apply hierarichal clustering over two party
computations which are using horizontally partitioned data.
Figure1. Two Party computations Model
DB DB DB DB
Randomizer Computing Engine
Query Front End Engine
Two partiesQuery Encoded query
Query Input
Response query

35
2.2 Partitioning Of Database
Partitioning of data base is the process in which we partition the data set in horizontal, vertical
and arbitrary. In the partitioning we apply several techniques to full fill the task of preserving the
data if we apply the perfect partitioning then we can easily make our approach good. First is
horizontally partitioning this means that the partitioning apply on the data set in which we are
deal with the data of a complete row we don’t have to worry about the column information a
complete row information comes in result.
In vertically partitioning the data is distributed in columns if any of the query come that means
that the information of a complete row comes in result this is good for if we are requesting for the
data of a single attribute.
The third partitioning approach is arbitrary partitioning that means that the horizontal and vertical
both type of approaches comes in this. When the query is requested from the database at that time
it is decided to apply either horizontal or vertical approach.
In all of the approaches the arbitrary approach is good but there is not much exploration in this
field but this approach is applied by using k-means clustering.
Table 1. Record of Students
S.No Name Branch Id
1 Ram CSE 234
2 Rohan CSE 235
3 Geeta CSE 223
4 Pooja ECE 342
Table 2. Vertically Partitioned
S.No Name
1 Ram
2 Rohan
3 Geeta
4 Pooja
Table 3.Horizontally Partitioned
S.No Name Branch Id
1 Ram CSE 234
2 Rohan CSE 235
In our approach we are going to use horizontally partitioned database hence we have the data
which is organised by row [5]. If we are taking data by a single attribute we have the all
information about a single attribute in a same time hence this is good approach for the banks and
academics. For this partitioning we use hierarchical clustering and also we are using
agglomerative approach with some encryption techniques.

36
3. CLUSTERING
Clustering is an approach which is best if we are dealing with some sensitive data or information.
The privacy is the major issue because there are many chances that the information is leak. Hence
clustering is the most appropriate approach for making the privacy strong[6] [7] [8].
Clustering is mainly said that if we are clustering some data then we have to find the data which
is most similar in their properties hencethey are cluster in a single group. Each group is different
from the other groups either in size, number of objects and their dimension and also they have
different data types.
Figure 2. Clustering of Database
Clustering is a data mining technique which is unsupervised data analysis [9]. It offers advanced
and more abstracted view to the dataset which is complex to handle if we are using simple
techniques. There are many clustering based privacy techniques which are given by researchers.
Types of clustering are following:-
1. Hierarchical clustering.
2. K-means clustering.
3. Density based clustering.
4. Self-organised maps EM clustering.
These are basic types of clustering algorithm which are mainly used in these days in many
approaches.
4. HIERARCHICAL CLUSTERING
Hierarichal clustering is one of the clustering approaches [10]. In hierarchal clustering is mainly
divided in two methods which are following:-
1. Agglomerative approach.
2. Divisive approach.
4.1 Agglomerative Approach
Agglomerative approach is one of the hierarichal clustering approach which is applied over the
database. In agglomerative approach this make cluster of database from its bottom to its top
C11 C12
C13 C23
C232
222
C211
Database
Cluster 1 Cluster 2

37
hence it is also called as bottom up approach. This is the most commonly used approach in the
field of clustering of data sets.
Figure 3. Agglomerative Approach
4.2 Divisive Approach
In divisive approach hierarichal clustering is applied over the database from top to bottom. Hence
this approach is called as top to bottom approach.
Figure 4. Divisive Approach
5. PRIVACY PRESERVING CLUSTERING ALGORITHM
5.1 K-Means clustering.
In our approach we are using k-means clustering algorithm for partitioning of datasets. The
followings steps are followed:
• Let X={X1,X2,……Xn} are the data elements and v={v1, v2,…….vc} are the set of
cluster center.
• First randomly select ‘c’ cluster center.
• Calculate the distance between each of data element and cluster center.
• Assign the minimum distance of each element from the entire cluster center.
• These steps are repeated until all of the elements come to a cluster.
abcde b
a
c
d
e
A
B
C
D
E
Ab
de
abcde
cde

38
5.2 Euclidian Distance Matrix
Euclidian distance matrix is an nxn matrix which represents the spacing of n points in any
Euclidian space.
Let there are two party P1 and P2 these are distributing the database D. P1 have the distance
matrix of first nxk elements and P2 have another nxk+1.
Now both of the parties have two set of data and the k-cluster center and distance matrix.
Before sharing the distance matrix we have to apply encryption over the data. In our approach we
use two encryption algorithms which are following:
1. SHA1
2. MD5
These two algorithms are applied over the data of two parties. Each of the party uses a different
encryption technique. Hence it is hard to understand for each of the party about the original
dataset.
The main advantage of this approach is that one dataset is damage then that will not affect the
other dataset.
5.3 Privacy Preserving Hierarichal agglomerative clustering.
Input: P1 have his cluster center and distance matrix and P2 have its own cluster center and
distance
Matrix.
Output:
• Assign all of the elements to a cluster.
• P1 compute k-cluster center (c1, c2, c3….ck) from the first attributes.
• P2 compute in the same manner the left attributes (ck+1, ck+2 ….c2k).
• P1 and P2 compute their cluster center and distance matrix as MP1 and MP2 respectively.
• P1 and P2 randomly share their cluster center and distance matrix. We use permute share
algorithm for sharing this information between them.
• P1 and P2 make the all possible cluster from the existing cluster informationi.e. k2
cluster
will be formed.
• Make a closest cluster from each party.
• Find the minimum value of each row of X matrix to find closest cluster for each instance
that is if the ith
column have minimum value in jth
instance then that will become the
closest ith
cluster.
• Place each instance to its appropriate closest cluster.
• Merge k2
cluster to make the final k cluster.
5.4 Algorithm for closest cluster.
Input: Given distance matrix of P1 and distance matrix of P2.
Output: first assign the closest cluster for n instance. A matrix X( nxk2
) that holds the
distance between each pair of n points and the k2
cluster center.

39
• For a=1 to n
• 1=0
• For b= 1to k
• For c=1 to k
• 1=1+1
• Xa1= pab +qac
• end for
• end for
• end for
• Return X
6. EFFICIENCY AND PRIVACY ANALYSIS
P1 and P2 both compute k-cluster on their own data set and then they share it to each other after
encrypting the value of data in distance matrix. The encryption is done through two algorithm
which are SHA1 and MD5 which takes O(k) time for each party. In the next step computational
complexity for computing the distance matrix by each party is calculated as O(nk). In the next
step hierarichal k-clustering take O(k2
) time for computation. The computational complexity for
closest cluster is O(nk2
). In the last step for each instance run time is O(k2
).
So the total time complexity is O(nk2
)
.
Both of the party send or receive their k cluster independently but in an encrypted form. So the
information of the parties does not public to other including the opposite party. They share only
the distance matrix but this distance matrix is only the distance computed between the cluster
center and instance. So the information is not leak. After merging the final k-cluster center is
exposing to each other. Hence the privacy preserving by using hierarichal clustering algorithm
over two parties using horizontally partitioned data is secure and does not leak any information.
7. EXPERIMENTAL RESULT
In the given approach we take a small database of 500 students. There records have weight and
height. These records are distributed in four clusters by using k-means clustering algorithm. After
this on two clusters we apply SHA-1 and on the other we apply MD-5 algorithm for encryption of
the data or information that these cluster have. After encryption these are shared between these
two party P1 and P2. The overall approach is explained briefly above.
Here we give the comparison of our approach with some other techniques which are similar in
work but take more time in query processing. We take basically k-means clustering over
horizontally partitioned data , hierarichal clustering of two parties over vertically partitioned data
and k-means clustering over vertically partitioned data.

40
A brief description is given in figure that how the given approach is better than the existing
methods.
Figure 5. Some of data set on which hierarichal clustering is applied
Figure 6: After clustering the database is divided in two party and 6 clusters

41
Graph 1: Cluster formed in Party1
Graph 2: Cluster formed in Party 2
The above graph shows the representation of data after the original database is distributed in
between the two parties. Hence there are total six cluster are formed in our experiment. On these
two different set of cluster we use two different encryption algorithms. Both the parties are
unaware of the encryption technique of other. Hence the privacy of data is high.
Figure 5.Encryption over party 1 using SHA-1
Figure 6.Encryption over party 2 using MD5
73 65 59
75 67
61
77
60
6274
70
66
0%
50%
100%
cluster 1 cluster 2 cluster 3
Party 1
Series 5
Series 4
Series 3
Series 2
Series 1
58 71 66
59 72 68
61
75 65
59
76
0
100
200
300
400
Cluster 1 Cluster 2 Cluster 3
Party 2
Series 4
Series 3
Series 2
Series 1

Table 4. Comparison among different cluster approach
Types of
cluster
k-means
clustering over
HPD
No of
database
scanning
300
Running time
(sec)
3.6304
Fig
8. CONCLUSION
In this paper we analyse the privacy preserving
are various techniques used to solve the problems like adding noise or encryption data value. In
this paper a hierarichal clustering approach for horizontally
novel approach to secure data.
9. FUTURE RESEARCH W
The future research work can be to find solution for hierarichal clustering for multiparty which
can be apply over horizontal and vertically partitioned data. The
further enhancing for arbitrary partitioned data.
REFERENCES
[1] J. W. Han and M. Kamber, “Data Mining: Concepts and Techniques,” 2 nd Edition, China Machine
Press, Beijing, 2006.
[2] J. Vaidya and C. Clifton, “Privacy Preserving K
Proceedings of the ninth ACM SIGKDD international conference on Knowledge discovery and data
mining, Washington DC, USA, 2003, pp. 206
0
1
2
3
4
C1
Comparison among different cluster approach
clustering over
Hierarichal with
HPD
k-means
clustering over
VPD
Hierarichal with
VPD
150 300 150
1.342 3.6309 1.451
Figure 7.Execution time of each approach
In this paper we analyse the privacy preserving problems for horizontally partitioned data. There
this paper a hierarichal clustering approach for horizontally partitioned data for two
WORK
can be to find solution for hierarichal clustering for multiparty which
can be apply over horizontal and vertically partitioned data. The hierarichal clustering can be
for arbitrary partitioned data.
J. W. Han and M. Kamber, “Data Mining: Concepts and Techniques,” 2 nd Edition, China Machine
J. Vaidya and C. Clifton, “Privacy Preserving K-Means Clustering over Vertically Partitioned Data”
mining, Washington DC, USA, 2003, pp. 206-215.doi:10.1145/956750.956776
C2
C3
C4
K-means HPD
Hierarichal HPD
K-means VPD
Hierarichal VPD
42
Hierarichal with
problems for horizontally partitioned data. There
partitioned data for two parties is a
can be to find solution for hierarichal clustering for multiparty which
hierarichal clustering can be
J. W. Han and M. Kamber, “Data Mining: Concepts and Techniques,” 2 nd Edition, China Machine
Means Clustering over Vertically Partitioned Data”
means HPD
Hierarichal HPD
means VPD
Hierarichal VPD

43
[3] T. K. Yu, D. T. Lee, Shih-Ming Chang and Justin Zhan, “Multi-Party k-Means Clustering with
Privacy Consideration,” International Symposium on Parallel and Dis-tribute Processing with
Applications, IEEE Computer Society, 2010, pp. 200- 207.
[4] P. Bunn and R. Ostrovsky, “Secure Two-Party k-Means Clustering,” In Proceedings of the 14th ACM
Conference on Computer and Communications Security, 2007, pp. 486-497.
doi:10.1145/1315245.1315306
[5] J. S. Vaidya, “Privacy Preserving Data Mining over Vertically Partitioned Data,” Ph.D. Thesis,
Purdue University, 2004, pp. 1-149.
[6] V. ESTIVILL-CASTRO, Why so many clustering algorithms: A position paper, SIGKDD
Explorations Newsletter, 4 (2002), pp. 65–75.
[7] J. Vaidya and C. Clifton, “Privacy Preserving K-Means Clustering over Vertically Partitioned Data,”
mining, Washington DC, USA, 2003, pp. 206-215. doi:10.1145/956750.956776
[8] T. K. Yu, D. T. Lee, Shih-Ming Chang and Justin Zhan, “Multi-Party k-Means Clustering with
Privacy Consideration,” International Symposium on Parallel and Distributed Processing with
Applications, IEEE Computer Society, 2010, pp. 200- 207.
[9] G. Jagannathan and R. N. Wright, “Privacy Preserving Distributed k-Means Clustering over
Arbitrarily Partitioned Data,” Proceedings of the 11th ACM, SIGKDD International Conference on
Knowledge Discovery and Data Mining, USA, 2005, pp. 1-7.
[10] I.De and A. tripathy,(2104), a secure two party hierarchal clustering approach for vertically
partitioned dataset with accuracy measure , 2nd international symp. Vol-34 no-3 page no-153-162.

TWO PARTY HIERARICHAL CLUSTERING OVER HORIZONTALLY PARTITIONED DATA SET

More Related Content

What's hot (19)

Similar to TWO PARTY HIERARICHAL CLUSTERING OVER HORIZONTALLY PARTITIONED DATA SET (20)

Recently uploaded (20)

TWO PARTY HIERARICHAL CLUSTERING OVER HORIZONTALLY PARTITIONED DATA SET