International Journal of Computer Trends and Technology (IJCTT) volume 4 Issue10 Oct 2013

ISSN: 2231-2803 http://www.ijcttjournal.org Page3648

Efficient Data Clustering with Link Approach
1
Y. Sireesha,
2
CH. Srinivas,
3
K.C. Ravi Kumar.

1
PG Scholar, Department of Computer Science and Engineering,
Sridevi Womens Engineering College
Hyderabad, A.P, India

2
Associate Professor, Department of Computer Science and Engineering,
Sridevi Womens Engineering College
Hyderabad, A.P, India


3
Head of the Department, Department of Computer Science and Engineering,
Sridevi Womens Engineering College
Hyderabad, A.P, India

Abstract: Data clustering faces lots of
studies and researches and at last the results
being competitive to conventional algorithms,
even though using these techniques finally we
are getting an incomplete information. The
existed partitioned-information matrix contains
particular cluster-data point relations only, with
lot entries which are not recognized. The paper
explores researches that preferres this crisis
decomposes the efficiency of the clustering
result, and it contains a new link-based
approach, which increases the conventional
matrix by revealing the entries which are not
recognized based upon the common things
which are present both clusters and in ensemble.
Often, a perfect link-based algorithm is invented
and used for the underlying common
assessment. After all those, to gain the
maximum clustering outputs, a graph
partitioning technique is used for a weighted
bipartite graph that is formulated from the

refined matrix. Results on various real data sets
suggest that the proposed link-based method
mostly performs both conventional clustering
algorithms for categorical data and also most
common cluster ensemble techniques.
I. INTRODUCTION
To examine the data set we have different
approaches through get down to its structure data
clustering is the efficient way. Because of the
beneficial characteristics of clustering like
mining, machine learning and pattern recognition.
Clustering deals with the data to get stick with
similar ones. Those similar ones will get into
group or cluster. There are sort of algorithms for
clustering like k-means and PAM which are used
for clustering the numerical data, these are used to
International Journal of Computer Trends and Technology (IJCTT) volume 4 Issue10 Oct 2013
ISSN: 2231-2803 http://www.ijcttjournal.org Page3649

get the distance between feature vectors. The
drawback is this not get inherited directly for
clustering purpose on the categorical data,
where domain values are discrete and have no
ordering defined. As a result, many categorical
data clustering algorithms have been introduced
in recent years, with applications to interesting
domains such as protein interaction data. The
initial method was developed by making use of
Gowers similarity coefficient. Following that,
the k-modes algorithm in extended the
conventional k-means with a simple matching
dissimilarity measure and a frequency-based
method to update centroids.

Fig 1. The basic process of cluster ensembles.
As a single-pass algorithm, Squeezer makes use
of a prespecified similarity threshold to
determine which of the existing clusters to
which a data point under examination is
assigned. LIMBO is a hierarchical clustering
algorithm that uses the Information Bottleneck
(IB) framework to define a distance measure for
categorical tuples. The concepts of evolutionary
computing and genetic algorithm have also been
adopted by a partitioning method for categorical
data, i.e., GAClust. Cobweb is a model-based
method primarily exploited for categorical data
sets. Different graph models have also been
investigated by the STIRR, ROCK, and CLICK
techniques. In addition, several density-based
algorithms have also been devised for such
purpose, for instance, CACTUS, COOLCAT, and
CLOPE. Although, a large number of algorithms
have been introduced for clustering categorical
data, the No Free Lunch theorem suggests1 there
is no single clustering algorithm that performs
best for all data sets and can discover all types of
cluster shapes and structures presented in data.
Each algorithm has its own strengths and
weaknesses. For a particular data set, different
algorithms, or even the same algorithm with
different parameters, usually provide distinct
solutions. Therefore, it is difficult for users to
decide which algorithm would be the proper
alternative for a given set of data. Recently,
cluster ensembles have emerged as an effective
solution that is able to overcome these limitations,
and improve the robustness as well as the quality
of clustering results. The main objective of cluster
ensembles is to combine different clustering
decisions in such a way as to achieve accuracy
superior to that of any individual clustering.
Examples of well-known ensemble methods are:
1. The feature-based approach that transforms the
problem of cluster ensembles to clustering
categorical data.
International Journal of Computer Trends and Technology (IJCTT) volume 4 Issue10 Oct 2013
ISSN: 2231-2803 http://www.ijcttjournal.org Page3650

2. The direct approach that finds the final
partition through relabeling the base clustering
results.
3. graph-based algorithms that employ a graph
partitioning methodology, and
4. The pairwise-similarity approach that makes
use of co-occurrence relations between data
points.
Despite notable success, these methods generate
the final data partition based on incomplete
information of a cluster ensemble. The
underlying ensemble-information matrix
presents only cluster-data point relationships
while completely ignores those among clusters.
As a result, the performance of existing cluster
ensemble techniques may consequently be
degraded as many matrix entries are left
unknown. This paper introduces a link-based
approach to refining the aforementioned matrix,
giving substantially less unknown entries. A
link-based similarity measure is exploited to
estimate unknown values from a link network of
clusters. This research uniquely bridges the gap
between the task of data clustering and that of
link analysis. It also enhances the capability of
ensemble methodology for categorical data,
which has not received much attention in the
literature. In addition to the problem of
clustering categorical data that is investigated
herein, the proposed framework is generic such
that it can also be effectively applied to other data
types.
II. DISCUSSION
The difficulty of categorical data analysis is
characterized by the fact that there is no inherent
distance (or similarity) between attribute values.
The RM matrix that is generated within the LCE
approach allows such measure between values of
the same attribute to be systematically quantified.
The concept of link analysis [34], [35], [36] is
uniquely applied to discover the similarity among
attribute values, which are modeled as vertices in
an undirected graph. In particular, two vertices are
similar if the neighboring contexts in which they
appear are similar. In other words, their similarity
is justified upon values of other attributes with
which they co-occur. While the LCE methodology
is novel for the problem of cluster ensemble, the
concept of defining similarity among attribute
values (especially with the case of direct
ensemble, Type-I) has been analogously adopted
by several categorical data clustering algorithms.
Initially, the problem of defining a context-based
similarity measure has been investigated in [61]
and [62]. In particular, an iterative algorithm,
called Iterated Contextual Distances (ICD), is
introduced to compute the proximity between two
values. Similar to LCE, the underlying distance
metric is based on the occurrence statistics of
attribute values. The WTQ algorithm is
summarized below
International Journal of Computer Trends and Technology (IJCTT) volume 4 Issue10 Oct 2013
ISSN: 2231-2803 http://www.ijcttjournal.org Page3651


However, the fundamental information model
that is used by ICD and LCE to capture the
associations between data points and attribute
values are notably different: a sequential
probabilistic chain and a link network for ICD
and LCE, respectively. Note that LCE makes
use of WTQ that is a single-pass similarity
algorithm, while ICD requires the chain model
to be randomly initialized and iteratively
updated to a fixed point.
III. SYSTEM DEVELOPMENT
The link based approach will be done in
following steps
K-Means with Euclidian Distance (1)
Transform categorical (text) into
numerical value
On Numerical data apply K-Means with
Euclidian distance measure.
The outcome of the 1 is {C11, C12,
C13 .. C1n} where C1 is the cluster 1
with 1 clustering algorithm.
K-Means with cosine similarity (2)
Consider numerical values are categorical
data
On categorical data apply (2) for
clustering.
The outcome of the 2 is {C21, C22,
C23.. C2n}
K-Means with Jaccords Coefficient (3)
Consider numerical values as categorical
data
On categorical data, apply (3) for
clustering
The outcome of the 3 is {C31, C32,
C33, C34 C3n}
Cluster Ensemble
Direct Ensemble
Full Space Ensemble
Subspace Ensemble
Generating Refined Matrix (RM)
Prepare RM as a matrix where each record
of the original dataset D as {x1, x2,
x3 xm}

International Journal of Computer Trends and Technology (IJCTT) volume 4 Issue10 Oct 2013
ISSN: 2231-2803 http://www.ijcttjournal.org Page3652


Weighted Triple Quality (WTQ)
Wxy = Lx Ly
Lx U Ly
Apply WTQ on RM, The output of the module
is refined clusters.
IV. RELATED WORK
Despite pursuing an objective analogous to that
of the LCE approach, several categorical data
clustering methods have been developed using
different mechanisms to specify a distance
between attribute values: STIRR, ROCK, and
CACTUS, for instance. STIRR is an iterative
algorithm based on nonlinear dynamical
systems. A database is encoded into a graph
structure, where each weighted node stands for a
specific attribute value. STIRR iteratively
updates the weight configuration until a stable
point (called basin) is reached. This is achieved
using a user-defined combiner function to
estimate a node weight from those of others that
associate to the same data records. Unlike LCE,
the similarity between any node pair cannot be
explicitly measured here. In fact, STIRR only
divides nodes of each attribute into two groups
(one with large positive weights and the other
with small negative weights) that correspond to
projections of clusters on the attribute. Yet, the
post processing required to generate the actual
clusters is nontrivial and not addressed in the
original work. While LCE is generally robust to
parameter settings, it is hard to analyze the
stability of the STIRR system for any useful
combiner function [63]. Rigorous experimentation
and fine tuning of parameters are needed for the
generation of a meaningful clustering [64]. ROCK
[14] makes use of a link graph, in which nodes
and links represent data points (or tuples) and
their similarity, respectively. Two tuples are
similar if they shared a large number of attribute
values. Note that the link connecting two nodes is
included only when the corresponding similarity
exceeds a user-defined threshold. With tuples
being initially regarded as singleton clusters,
ROCK merges clusters in an agglomerative
hierarchical fashion, while optimizing a cluster
quality that is defined in terms of the number of
links across clusters. Note that the graph models
International Journal of Computer Trends and Technology (IJCTT) volume 4 Issue10 Oct 2013
ISSN: 2231-2803 http://www.ijcttjournal.org Page3653

used by ROCK and LCE are dissimilarthe
graph of data points and that of attribute values
(or clusters), respectively. Since the number of
data points is normally greater than that of
attribute values, ROCK is less efficient than
LCE. As a result, it is unsuitable for large data
sets. Also, the selection of a smooth function
that is used to estimate a cluster quality is a
delicate and difficult task for average users.
CACTUS also relies on the co-occurrence
among attribute values. In essence, two attribute
values are strongly connected if their support
(i.e., the proportion of tuples in which the values
co-occur) exceeds a prespecified value. By
extending this concept to all attributes,
CACTUS searches for the distinguishing sets,
which are attribute value sets that uniquely
occur within only one cluster. These sets
correspond to cluster projections that can be
combined to formulate the final clusters. Unlike
LCE, the underlying problem is not designed
using a graph based concept. It is also
noteworthy that CACTUS and its recent
extension assume each cluster to be identified
by a set of attribute values that occur in no other
cluster. While such conjecture may hold true for
some data sets, it is unnatural and unnecessary
for the clustering process. This rigid constraint
is not implemented by the LCE method. Besides
these approaches, traditional categorical data
analysis also utilizes the market-basket
numerical representation of the nominal data
matrix. This transformed matrix is similar to the
BM, which has been refined to the RM
counterpart by LCE. A similar attempt in
identifies the connection between category
utility of the conceptual clustering (Cobweb) and
the classical objective function of k-means.

Fig 2. An example of a cluster network, where each edge is
marked with its weight.

As a result, the so-called market-basket matrix
used by the former is transformed to a variation
that can be efficiently utilized by the latter. The
intuitions of creating this rescaled matrix and the
RM are fairly similar. However, the methods used
to generate them are totally different. LCE
discovers unknown entries (i.e., 0) in the
original BM from known entries (1), which are
preserved and left unchanged. On the other hand,
the method in maps the attribute-value-specific
1 and 0 entries to the unique standardized
values. Unlike the RM, this matrix does not
International Journal of Computer Trends and Technology (IJCTT) volume 4 Issue10 Oct 2013
ISSN: 2231-2803 http://www.ijcttjournal.org Page3654

conserve the known fact (1 entries), whose
values are now different from one to another
attribute. Despite the fact that many clustering
algorithms and LCE are developed with the
capability of comparing attribute values in
mind, they achieve the desired metric
differently, using specific information models.
LCE uniquely and explicitly models the
underlying problem as the evaluation of link-
based similarity among graph vertices, which
stand for specific attribute values (for Type-I
ensemble) or generated clusters (for Type-II and
Type-III). The resulting system is more efficient
and robust, as compared to other clustering
techniques emphasized thus far. In addition to
SPEC, many other classical clustering
techniques, k-means and PAM among others,
can be directly used to generate the final data
partition from the proposed RM. The LCE
framework is generic such that it can be adopted
for analyzing other types of data.
V. CONCLUSION
This paper presents a novel, highly effective
link-based cluster ensemble approach to
categorical data clustering. It transforms the
original categorical data matrix to an
information-preserving numerical variation
(RM), to which an effective graph partitioning
technique can be directly applied. The problem
of constructing the RM is efficiently resolved by
the similarity among categorical labels (or
clusters), using the Weighted Triple-Quality
similarity algorithm. The empirical study, with
different ensemble types, validity measures, and
data sets, suggests that the proposed link-based
method usually achieves superior clustering
results compared to those of the traditional
categorical data algorithms and benchmark cluster
ensemble techniques. The prominent future work
includes an extensive study regarding the
behavior of other link-based similarity measures
within this problem context. Also, the new
method will be applied to specific domains,
including tourism and medical data sets.
REFERENCES
[1] D.S. Hochbaum and D.B. Shmoys, A Best
Possible Heuristic for the K-Center Problem,
Math. of Operational Research, vol. 10, no. 2, pp.
180-184, 1985.
[2] L. Kaufman and P.J. Rousseeuw, Finding
Groups in Data: An Introduction to Cluster
Analysis. Wiley Publishers, 1990.
[3] A.K. Jain and R.C. Dubes, Algorithms for
Clustering. Prentice-Hall, 1998.
[4] P. Zhang, X. Wang, and P.X. Song,
Clustering Categorical Data Based on Distance
Vectors, The J. Am. Statistical Assoc., vol. 101,
no. 473, pp. 355-367, 2006.
[5] J. Grambeier and A. Rudolph, Techniques of
Cluster Algorithms in Data Mining, Data Mining
and Knowledge Discovery, vol. 6, pp. 303-360,
2002.
International Journal of Computer Trends and Technology (IJCTT) volume 4 Issue10 Oct 2013
ISSN: 2231-2803 http://www.ijcttjournal.org Page3655

[6] K.C. Gowda and E. Diday, Symbolic
Clustering Using a New Dissimilarity Measure,
Pattern Recognition, vol. 24, no. 6, pp. 567-
578, 1991.
[7] J.C. Gower, A General Coefficient of
Similarity and Some of Its Properties,
Biometrics, vol. 27, pp. 857-871, 1971.
[8] Z. Huang, Extensions to the K-Means
Algorithm for Clustering Large Data Sets with
Categorical Values, Data Mining and
Knowledge Discovery, vol. 2, pp. 283-304,
1998.
[9] Z. He, X. Xu, and S. Deng, Squeezer: An
Efficient Algorithm for Clustering Categorical
Data, J. Computer Science and Technology,
vol. 17, no. 5, pp. 611-624, 2002.
[10] P. Andritsos and V. Tzerpos, Information-
Theoretic Software Clustering, IEEE Trans.
Software Eng., vol. 31, no. 2, pp. 150-165, Feb.
2005.
[11] D. Cristofor and D. Simovici, Finding
Median Partitions Using Information-
Theoretical-Based Genetic Algorithms, J.
Universal Computer Science, vol. 8, no. 2, pp.
153-172, 2002.
[12] D.H. Fisher, Knowledge Acquisition via
Incremental Conceptual Clustering, Machine
Learning, vol. 2, pp. 139-172, 1987.