International Journal of Scientific & Engineering Research, Volume 7, Issue 2, February-2016

ISSN 2229-5518 247

Importance of Clustering in Data Mining

Prof. M. A. Deshmukh Prof. R. A. Gulhane

Assistant professor in Computer Science and Assistant Professor in Computer Science and Engineering
Engineering Prof. Ram Meghe Institute of Technology & Research,
Prof. Ram Meghe Institute of Technology & Research, Badnera. Maharashtra,India
Badnera.Maharashtra,India [email protected]
[email protected]

Abstract— Cluster analysis groups objects clusters. Clustering is important in data analysis and data
(observations, events) based on the information found in mining applications[1]. It is the task of grouping a set of
the data describing the objects or their relationships. The objects so that objects in the same group are more
similar to each other than to those in other groups. A
goal is that the objects in a group will be similar (or
good clustering algorithm is able to identity clusters
related) to one other and different from (or unrelated to) irrespective of their shapes. The stages involved in
the objects in other groups. The greater the similarity (or clustering algorithm are as follows,
homogeneity) within a group, and the greater the
Raw data
difference between groups, the ― better‖ or more distinct
the clustering. Data mining is the process of analysing

IJSER
data from different viewpoints and summerising it into
useful information. Data mining is one of the top clustering
algorithms
research areas in recent days. Cluster analysis in data
mining is an important research field it has its own
unique position in a large number of data analysis and
processing. clusters of
data

II. Literature Review

I. Introduction
Cluster analysis itself is not one specific
Cluster analysis or clustering is the task of grouping a algorithm, but the general task to be solved. It can be
set of objects in such a way that objects in the same achieved by various algorithms that differ significantly
group (called a cluster) are more similar (in some sense in their notion of what constitutes a cluster and how to
or another) to each other than to those in other groups efficiently find them. Popular notions of clusters include
(clusters). It is a main task of exploratory data mining, groups with small distances among the cluster members,
and a common technique for statistical data analysis, dense areas of the data space, intervals or particular
used in many fields, including machine learning, pattern statistical distributions. Clustering can therefore be
recognition, image analysis, information retrieval, and formulated as a multi-objective optimization problem.
bioinformatics. The appropriate clustering algorithm and parameter
Data mining is the process of analysing data from settings (including values such as the distance function
different perspectives and summarizing it into useful to use, a density threshold or the number of expected
information. Data mining involves the anomaly clusters) depend on the individual data set and intended
detection, association rule learning, classification, use of the results. Cluster analysis as such is not an
regression, summarization and clustering. In this paper, automatic task, but an iterative process of knowledge
clustering analysis is done. A cluster is a collection of discovery or interactive multi-objective optimization that
data objects that are similar to one another within the involves trial and failure. It will often be necessary to
same cluster and are dissimilar to the objects in other

IJSER © 2016
http://www.ijser.org
International Journal of Scientific & Engineering Research, Volume 7, Issue 2, February-2016
ISSN 2229-5518 248

modify data preprocessing and model parameters until R, then the distance of the new cluster, R, to an existing
the result achieves the desired properties….. cluster, Q, is a linear function of
Besides the term clustering, there are a number of the distances of Q from the original clusters A and B.
terms with similar meanings, including automatic Any hierarchical technique that can be phrased in this
classification, numerical taxonomy, The subtle way does not need the
differences are often in the usage of the results: while in original points, only the proximity matrix, which is
data mining, the resulting groups are the matter of updated as clustering occurs.
interest, in automatic classification the resulting However, while a general formula is nice, it is often
discriminative power is of interest. This often leads to easier to understand the different
misunderstandings between researchers coming from the hierarchical methods by looking directly at the definition
fields of data mining and machine learning, since they of cluster distance that each
use the same terms and often the same algorithms, but method uses, and that is the approach that we shall take
have different goal. here. [DJ88] and [KR90] both
give a table that describes each method in terms of the
Why clustering? Lance-Williams formula
.
• Organizing data into clusters shows internal Mutual Nearest Neighbor Clustering
structure of the data
– Ex. Clusty and clustering genes above Mutual nearest neighbor clustering is described in
• Sometimes the partitioning is the goal [GK77]. It is based on the idea
– Ex. Market segmentation of the ― mutual neighborhood value (mnv)‖ of two points,
• Prepare for other AI techniques which is the sum of the ranks of

IJSER
– Ex. Summarize news (cluster and then find centroid) the two points in each other’s sorted nearest-neighbor
• Techniques for clustering is useful in knowledge lists. Two points are then said to
discovery in data be mutual nearest neighbors if they are the closest pair of
– Ex. Underlying rules, reoccurring patterns, topics, etc. points with that mnv.
Clusters are built up by starting with points as singleton
Methods: clusters and then merging
the closest pair of clusters, where close is defined in
Basic Agglomerative Hierarchical Clustering terms of the mnv. The mnv between
Algorithm two clusters is the maximum mnv between any pair of
1) Compute the proximity graph, if necessary. points in the combined cluster. If
(Sometimes the proximity graph is all that is available.) there are ties in mnv between pairs of clusters, they are
2) Merge the closest (most similar) two clusters. resolved by looking at the original
3) Update the proximity matrix to reflect the proximity distances between points. Thus, the algorithm for mutual
between the new cluster and the nearest neighbor clustering
original clusters. works in the following way.
4) Repeat steps 3 and 4 until only a single cluster a) First the k-nearest neighbors of all points are
remains. found. In graph terms this can be
The key step of the previous algorithm is the calculation regarded as breaking all but the k strongest links from a
of the proximity between point to other points in the
two clusters, and this is where the various agglomerative proximity graph.
hierarchical techniques differ. b) For each of the k points in a particular point’s k-
Any of the cluster proximities that we discuss in this nearest neighbor list, calculate
section can be viewed as a choice of the mnv value for the two points. It can happen that a
different parameters (in the Lance-Williams formula) for point is in one point’s knearest
the proximity between clusters neighbor list, but not vice-versa. In that case, set the mnv
Q and R, where R is formed by merging clusters A and B. value to some value
p(R, Q) = A p(A, Q) + B p(B, Q) + p(A, Q) + larger than 2k.
| p(A, Q) – p(B, Q) | c) Merge the pair of clusters having the lowest mnv
In words, this formula says that after you merge clusters (and the lowest distance in case
A and B to form cluster of ties).

IJSER © 2016
http://www.ijser.org
International Journal of Scientific & Engineering Research, Volume 7, Issue 2, February-2016
ISSN 2229-5518 249

d) Repeat step (c) until the desired number of clusters

complexity is for agglomerative clustering and
is reached or until the only
clusters remaining cannot be merged. The latter case for divisive clustering, which makes them too
will occur when no points in slow for large data sets. For some special cases, optimal
different clusters are k-nearest neighbors of each other.
The mutual nearest neighbor technique has behavior efficient methods (of complexity ) are known:
similar to the shared nearest SLINK for single-linkage and CLINK for complete-
neighbor technique in that it can handle clusters of linkage clustering. In the data mining community these
varying density, size, and shape. methods are recognized as a theoretical foundation of
However, it is basically hierarchical in nature while the cluster analysis, but often considered obsolete. They did
shared nearest neighbor approach however provide inspiration for many later methods such
is partitional in nature. as density based clustering.
Connectivity based clustering, also known as  Linkage clustering examples
hierarchical clustering, is based on the core idea of
objects being more related to nearby objects than to
objects farther away. These algorithms connect "objects"
to form "clusters" based on their distance. A cluster can
be described largely by the maximum distance needed to
connect parts of the cluster. At different distances,
different clusters will form, which can be represented
using a dendrogram, which explains where the common
name "hierarchical clustering" comes from: these

IJSER
algorithms do not provide a single partitioning of the
data set, but instead provide an extensive hierarchy of
clusters that merge with each other at certain distances.
In a dendrogram, the y-axis marks the distance at which 
the clusters merge, while the objects are placed along the
Single-linkage on Gaussian data. At 35 clusters,
x-axis such that the clusters don't mix.
the biggest cluster starts fragmenting into
Connectivity based clustering is a whole family of smaller parts, while before it was still connected
methods that differ by the way distances are computed. to the second largest due to the single-link
Apart from the usual choice of distance functions, the effect.
user also needs to decide on the linkage criterion (since a
cluster consists of multiple objects, there are multiple
candidates to compute the distance to) to use. Popular
choices are known as single-linkage clustering (the
minimum of object distances), complete linkage
clustering (the maximum of object distances) or
UPGMA ("Unweighted Pair Group Method with
Arithmetic Mean", also known as average linkage
clustering). Furthermore, hierarchical clustering can be
agglomerative (starting with single elements and
aggregating them into clusters) or divisive (starting with
the complete data set and dividing it into partitions).

These methods will not produce a unique partitioning of Single-linkage on density-based clusters. 20
the data set, but a hierarchy from which the user still clusters extracted, most of which contain single
needs to choose appropriate clusters. They are not very elements, since linkage clustering does not have
robust towards outliers, which will either show up as a notion of "noise".
additional clusters or even cause other clusters to merge
(known as "chaining phenomenon", in particular with
single-linkage clustering). In the general case, the

IJSER © 2016
http://www.ijser.org
International Journal of Scientific & Engineering Research, Volume 7, Issue 2, February-2016
ISSN 2229-5518 250

General Types of Clusters segmentation, target customer orientation, performance

1. Well-separated clusters assessment, biological species etc.
A cluster is a set of points so that any point in a cluster is
nearest (or more similar) to every other point in the
cluster as compared to any other point that is not in the Some Relative Applications
cluster.
2. Centre-based clusters The cluster analysis has been applied to many occasions.
A cluster is a set of objects such that an object in a For example, in commercial, cluster analysis was used to
cluster is nearest (more similar) to the ― centre‖ of a find the different customer groups, and summarize
cluster, than to the centre of any other cluster [2]. The different customer group characteristics through the
centre of a cluster is often a centroid. buying habits; in biotechnology, cluster analysis was
3. Contiguous clusters used to categorized animal and plant populations
A cluster is a set of points so that a point in a cluster is according to population and to obtain the latent structure
nearest (or more similar) to one or more other points in of knowledge; in geography, clustering can help
the cluster as compared to any point that is not in the biologists to determinate the relationship of the different
cluster. species and different geographical climate; in the
4. Density-based clusters banking sector, by using cluster analysis to bank
A cluster is a dense region of points, which is separated customers to refine a user group; in the insurance
by according to the low-density regions, from other industry, according to the type of residence, around the
regions that is of high density. business district, the geographical location, cluster
5. Shared Property or Conceptual Clusters analysis can be used to complete an automatic grouping
Finds clusters that share some common property or of regional real estate, to reduce the manpower cost and

IJSER
represent a particular concept. insurance company industry risk; in the Internet, cluster
analysis was used for document classification and
III. Analysis of Clustering Algorithm information retrieval etc.

Clustering is the main task of Data Mining and it is done V.conclusion

by the number of algorithms. The most commonly used
algorithms in Clustering are Hierarchical, Partitioning
and Grid based algorithms [1]. The overall goal of the data mining process is to
extract information from a large data set and transform it
1. Hierarchical Algorithms into an understandable form for further use. Clustering is
Hierarchical clustering is a method of cluster analysis important in data analysis and data mining applications.
which seeks to build a hierarchy of clusters. It is the It is the task of grouping a set of objects so that objects
connectivity based clustering algorithms. The in the same group are more similar to each other than to
hierarchical algorithms build clusters gradually. those in other groups (clusters). Clustering can be done
Hierarchical clustering generally fall into two types: In by the different no. of algorithms such as hierarchical,
hierarchical clustering, in single step, the data are not partitioning and grid algorithms. Hierarchical clustering
partitioned into a particular cluster [3]. is the connectivity based clustering. Partitioning is the
centroid based clustering; the value of k-mean is set. The
application of cluster analysis is more and more urgent;
IV. The Application of Cluster Analysis in Data
the requirements are also getting higher and higher. With
Mining
the development of modern technology, in the near
future, cluster areas will achieve a critical breakthrough.
The application of cluster analysis in data mining has
two main aspects: first, clustering analysis can be used as
a pre-processing step for the other algorithms such as
References
features and classification algorithm, and also can be
used for further correlation analysis. Second, it can be
[1] https://en.wikipedia.org/wiki/Cluster_analysis
used as a stand-alone tool in order to get the data
distribution, to observe each cluster features, then focus
[2] http://www.ise.bgu.ac.il/faculty/liorr/hbchap15.pdf
on a specific cluster for some further analysis [5].
Cluster analysis can be available in market

IJSER © 2016
http://www.ijser.org
International Journal of Scientific & Engineering Research, Volume 7, Issue 2, February-2016
ISSN 2229-5518 251

[3]
http://www.academia.edu/7764213/Analysis_and_Appli
cation_of_Clustering_Techniques_in_Data_Mining

[4] Yan-Ying Chen, An-Jung Cheng, ― Travel

Recommendation by Mining People Attributes and
Travel Group Types From Community-Contributed
Photos‖ IEEE Transactions on Multimedia, Vol. 15, No.
6, October 2013.

IJSER

IJSER © 2016
http://www.ijser.org