A Comparative Study Of Various Clustering Algorithms In Data Mining

K. Chitra et al, International Journal of Computer Science and Mobile Computing, Vol.6 Issue.8, August- 2017, pg. 109-115
© 2017, IJCSMC All Rights Reserved 109
Available Online at www.ijcsmc.com
International Journal of Computer Science and Mobile Computing
A Monthly Journal of Computer Science and Information Technology
ISSN 2320–088X
IMPACT FACTOR: 6.017
IJCSMC, Vol. 6, Issue. 8, August 2017, pg.109 – 115
A Comparative Study of Various
Clustering Algorithms in Data Mining
K. Chitra1
, Dr. D.Maheswari2
Research Scholar, School of Computer Studies, Rathnavel Subramaniam College of Arts and Science,
Sulur, Coimbatore1
Head, Research Coordinator, School of Computer Studies- PG, Rathnavel Subramaniam College of
Arts and Science, Sulur, Coimbatore2
E mail: chitra.k@rvsgroup.com 1
, maheswari@rvsgroup.com 2
Abstract --The purpose of the data mining technique is to mine information from a bulky data set and make it
into a reasonable form for supplementary purpose. Data mining can do by passing through various phases.
Mining can be done by using supervised and unsupervised learning. Clustering is a significant task in data
analysis and data mining applications. It is the task of arranging a set of objects so that objects in the
identical group are more related to each other than to those in other groups (clusters). The clustering is
unsupervised learning. Clustering algorithms can be classified into partition-based algorithms, hierarchical-
based algorithms, density-based algorithms and grid-based algorithms. This paper focuses on a keen study of
different clustering algorithms in data mining. A brief overview of various clustering algorithms is discussed.
Keywords: data mining, clustering, clustering algorithms, techniques
I. INTRODUCTION
Data mining refers to extracting information from large amounts of data, and transforming that
information into an understandable and meaningful structure for further use. Data mining is an essential step in
the process of knowledge discovery from data (or KDD). It helps to extract patterns and make hypothesis from
the raw data. Tasks in data mining include anomaly detection, association rule learning, classification,
regression, summarization and clustering [1].
In Data Mining the two types of learning sets are used, they are supervised learning and unsupervised learning.
a) Supervised Learning
In supervised training, data includes together the input and the preferred results. It is the rapid and
perfect technique. The accurate results are recognized and are given in inputs to the model through the learning
procedure. Supervised models are neural network, Multilayer Perceptron and Decision trees.

b) Unsupervised Learning
The unsupervised model is not provided with the accurate results during the training. This can be used
to cluster the input information in classes on the basis of their statistical properties only. Unsupervised models
are for dissimilar types of clustering, distances and normalization, k-means, self organizing maps.
II. CLUSTERING
Clustering is a major task in data analysis and data mining applications. It is the assignment of
combination a set of objects so that objects in the identical group are more related to each other than to those in
other groups. Cluster is an ordered list of data which have the familiar characteristics. Cluster analysis can be
done by finding similarities between data according to the characteristics found in the data and grouping similar
data objects into clusters. Clustering is an unsupervised learning process. Clustering has an extensive and
prosperous record in a range of scientific fields in the vein of image segmentation, information retrieval and web
data mining.
Figure 1: Stages of Clustering [15]
III. CLUSTERING ALGORITHMS
Clustering algorithms can be categorized into partition-based algorithms hierarchical-based algorithms,
density-based algorithms and grid-based algorithms. These methods vary in (i) the procedures used for
measuring the similarity (within and between clusters) (ii) the use of thresholds in constructing clusters (iii) the
manner of clustering, that is, whether they allow objects to belong to strictly to one cluster or can belong to
more clusters in different degrees and the structure of the algorithm. Irrespective of the method used, the
resulting cluster structure is used as a result in itself, for inspection by a user, or to support retrieval of objects
[5].
A. Partitioning Algorithms
Partitioning clustering algorithm split the data points into k division, where each division represent a
cluster and k<=n, where n is the number of data points. Partitioning methods are based on the idea that a cluster
can be represented by a centre point. The cluster must exhibit two properties, they are (a) each collection should
have at least one object (b) every object should belong to accurately one collection. The main drawback of this
algorithm is whenever a point is close to the center of another cluster; it gives poor outcome due to overlapping
of data points [4]. It uses a number of greedy heuristics schemes of iterative optimization.
There are many methods of partitioning clustering; they are K-Means, K-Medoids Method, PAM
(Partitioning around Medoids), and CLARA (Clustering Large Applications)[8].
1) K-Means algorithm: In this algorithm a cluster is represented by its centroid, which is a mean (average pt.) of
points within a cluster [3]. This works efficiently only with numerical attributes. And it can be negatively
affected by a single outlier. The k-means algorithm is the most popular clustering tool that is used in scientific
and industrial applications. The technique aims to partition “n” observations into k clusters in which every
observation belongs to the cluster with the nearby mean.
The basic algorithm is very simple
1. Select K points as initial centroids.
2. Repeat.
3. Form K clusters by assigning each point to its closest centroid.
4. Re-compute the centroid of each cluster until centroid does not change.

The k-means algorithm has the following significant properties:
1. It is effective in dealing out huge data sets.
2. It frequently terminates at a local optimum.
3. It works just on numeric values.
4. The clusters have convex shapes [11].
Disadvantages of k-means
1. Generally terminates at the local optimum, and not the global optimum.
2. Can only be used when the mean is defined and therefore requires specifying k, the number of
clusters, in advance.
2) K-Medoids Algorithm: In this algorithm we utilize the actual entity to represent the cluster, using one
representative entity per cluster. Clusters are generated by points which are close to respective methods. The
partitioning is made based on minimizing the sum of the dissimilarities among every object and its cluster
representative [13].
PAM: Like all partitioning methods, PAM works in an iterative, greedy way. The initial representative objects
are chosen randomly, and it is considered whether replacing the representative objects by non-representative
objects would improve the quality of clustering. The representative objects are replaced with other objects and it
continues until the quality cannot be enhanced further. PAM searches for the best k-medoids among a given data
set. The time complexity of PAM is O(k(n-k)2) [13]. For large values of n and k, this computation becomes
even more costly than the k-means method.
Algorithm [13]:
a) Randomly choose k objects in D as the first representative objects or seeds.
b) Repeat
i) Allocate each lasting object to the cluster with the nearby representative object
ii) Arbitrarily choose a non-representative object, orandom
iii) Calculate the total cost, S, of swapping representative objects oj with orandom
iv) If S<0 then swap oj with orandom to form the new set of k representative objects.
c) Until no change
CLARA: To deal with huge data sets, a sampling based technique, called CLARA (Clustering Large
Applications) and an improved version which is based on randomized search called CLARANS (Clustering
Large Applications based upon Randomized Search) can be used[14].
CLARA uses 5 samples, each with 40+2k points, each of which are then subjected to PAM, which
computes the best medoids from the sample [2]. A large sample usually works well when all the objects have
equal probability of getting selected. The complexity of computing medoids from a random sample is
O(ks2+k(n-k)), Where s is the size of the sample [13].
CLARA cannot find a good clustering if any of the best sampled medoids is far from the best k-
medoids.
Pros and Cons of Partitioning Algorithm: It is simple to understand and implement. It takes less time to execute
as compared to other techniques. The drawback of this algorithm is the user has to provide pre-determined value
of k and it produces spherical shaped clusters. It cannot handle with noisy data objects
B. Hierarchical Algorithm
This algorithm partitions the related dataset by constructing a hierarchy of clusters. It uses the distance
matrix criteria for clustering the data. It constructs clusters step by step. In hierarchical clustering, in single step,
the data are not partitioned into a particular cluster. It takes a sequence of partitions, which may run from a

single cluster containing all objects to „n‟ clusters each containing a single object. Hierarchical Clustering is
classified as
1. Agglomerative Nesting
2. Divisive Analysis
1) Agglomerative Nesting: It is also known as AGNES. It is bottom-up approach. This method construct the tree
of clusters i.e. nodes. The criteria used in this method for clustering the data is min distance, max distance, avg
distance, center distance. The steps of this method are:
(1) Initially all the objects are clusters i.e. leaf.
(2) It recursively merges the nodes (clusters) that have the maximum similarity between them.
(3) At the end of the process all the nodes belong to the same cluster i.e. known as the root of the tree
structure.
BIRCH (Balanced Iterative Reducing and Clustering using Hierarchies): BIRCH is an agglomerative
hierarchical based clustering algorithm. It is used for clustering large amounts of data. It is based on the notion
of a clustering feature (CF) and a CF tree. A CF tree is a height-balanced tree. Leaf nodes consist of a sequence
of clustering features, where each clustering feature represents points that have already been scanned. It is
mainly used when a small number of I/O operations are needed. BIRCH uses a multi clustering technique,
wherein a basic and good clustering is produced as a result of the first scan, and additional scans can be used to
further improve the quality of clustering [13]. The time complexity of BIRCH is O(n) where n is number of
clusters [13].
CHAMELEON: CHAMELEON is an agglomerative hierarchical clustering technique where, unlike other
algorithms which use static model of clustering, CHAMELEON uses dynamic modeling to merge the clusters. It
does not need users to provide information, but adjusts the clusters to be merged automatically according to
their intrinsic properties. It has a complexity of O(Nm+Nlog(N)+m2log(m)), where N is the number of data
points and m is the number of partitions [16]. CHAMELEON uses the interconnectivity and compactness of the
clusters to find out the similarity between them. It can be used to find clusters of varying densities [12].
However, the processing time for high-dimensional data may go up to O(n2).
2) Devise Analysis: It is also known as DIANA. It is top-down approach. It is introduced in Kaufmann and
Rousseeuw (1990). It is the inverse of the agglomerative method. Starting from the root node (cluster) step by
step each node forms the cluster (leaf) on its own. It is implemented in statistical analysis packages, e.g., plus.
Figure 2: Representation of AGNES and DIANA
Pros and Cons of hierarchical clustering [2]
It has embedded flexibility with regard to the level of granularity and it is easy to handle any forms of
similarity or distance and applicable to any attributes type. The drawback of this algorithm is most hierarchical
algorithm does not revisit once constructed clusters with the purpose of improvement.

C. Density Based Algorithms
Density based algorithms locate the cluster according to the regions which grow with high density. It is
the one-scan algorithms. The first approach called the density-based connectivity clustering pins density to a
training data point. It is able to find the arbitrary shaped clusters and handle noise. Representative algorithms
include DBSCAN, GDBSCAN, OPTICS (Ordering Points to Identify the Clustering Structure), and DBCLASD.
The second approach pins density to a point in the attribute space and is called Density Functions. It includes the
algorithm DENCLUE.
1) DBSCAN (Density Based Spatial Clustering of Applications with Noise): The density based algorithm
DBSCAN is commonly known. The Eps and the Minpts are the two parameters of the DBSCAN [6]. The basic
idea of DBSCAN algorithm is that a neighborhood around a point of a given radius (ε) must contain at least
minimum number of points (MinPts) [6]. The steps of this method are:
(1) Randomly select a point t
(2) Recover all density-reachable points from t wrt Eps and MinPts.
(3) Cluster is created, if t is a core point
(4) DBSCAN visits the next point of the database. If t is a border point, and no points are density-reachable
from t.
(5) Continue the procedure until all of the points have been processed.
2) DENCLUE (Density-based Clustering): Denclue is a clustering method that depends upon density
distribution function. DENCLUE uses a gradient hill-climbing technique for finding local maxima of density
functions [10]. These local maxima are called density attractors, and only those local maxima whose kernel
density approximation is greater than the noise threshold are considered in the cluster. [13]
Pros and Cons of Density-Based Algorithm: The advantage of this algorithm is it does not require a-priori
specification and able to handle large amount of noise in dataset. It fails in case of neck type of dataset and it
does not work well in case of high dimensionality data [13].
D. Grid Density Based Algorithms
Grid Density based clustering is concerned with the value space that surrounds the data points not with
the data points. This algorithm uses the multi resolution grid data structure and use dense grids to form clusters.
Grid Density based algorithms require the users to specify a grid size or the density threshold, the problem here
arise is that how to choose the grid size or density thresholds. To overcome this problem, a technique of
adaptive grids are proposed that automatically determines the size of grids based on the data distribution and
does not require the user to specify any parameter like grid size or the density threshold. The grid- based
clustering algorithms are STING, Wave Cluster, and CLIQUE
1) STING (Statistical Information Grid approach): This approach breaks the available space of objects into cells
of rectangular shapes in a hierarchy. It follows the top down approach and the hierarchy of the cells can contain
multiple levels corresponding to multiple resolutions [12]. The statistical information like the mean, maximum
and minimum values of the attributes is precomputed and stored as statistical parameters, and is used for query
processing and other data analysis tasks. The statistical parameters for higher level cells can be computed from
the parameters for lower level cells. The complexity is O (K) where k denotes the total count of cells in last tier.
2) CLIQUE: Clique is a grid-based method that finds density-based clusters in subspaces. CLIQUE performs
clustering in two steps. In the first step, CLIQUE partitions each dimension into non-overlapping rectangular
units, thereby partitioning the entire space of data objects into cells. At the same time it identifies the dense cells
in all the subspaces. The unit is dense when the fraction of total data points exceeds the input model parameter.
In the second step, CLIQUE uses these dense cells to form clusters, which can be arbitrary.

Pros and cons of Grid Based Algorithm [7]: The advantage of this algorithm is its quick processing time and
this is typically independent of the number of data objects. The Disadvantage is it depends on only the number
of cells in each dimension in the quantized space.
IV. CONCLUSION
The objective of the data mining technique is to mine information from a large data set and make it into
a reasonable form for the supplementary purpose. Clustering is a significant task in data analysis and data
mining applications. It is the task of arranging a set of objects so that objects in the identical group are more
related to each other than to those in other groups (clusters). Clustering algorithms can be classified into
partition-based algorithms, hierarchical-based algorithms, density-based algorithms and grid-based algorithms.
Under partitioning method, a brief description of k-means and k-medoids algorithms have been studied. In
hierarchical clustering, the BIRCH and CHAMELEON algorithms have been described. The DBSCAN and
DENCLUE algorithms under the density based methods have been studied. Finally, under grid-based clustering
method, the STING and CLIQUE algorithms have been described. The challenge with clustering analysis is
mainly that different clustering techniques give substantially different results on the same data. Moreover, there
is no algorithm present which gives all the desired outputs. Because of this, there is extensive research being
carried out in „ensembles‟ of clustering algorithms, i.e. multiple clustering techniques done on a single dataset.
Table 1. Comparison of the features of various clustering algorithms [9]

REFERENCES
[1]. K. Kameshwaran and K. Malarvizhi, “Survey on Clustering Techniques in Data Mining”, International
Journal of Computer Science and Information Technologies (0975-9646), Vol. 5(2), 2014
[2]. Pradeep Rai and Shubha Singh (2010) A Survey of Clustering Techniques, International Journal of
Computer Applications (0975 – 8887) Vol 7– No.12, pp. 1-5
[3]. V.Kavitha, M.Punithavalli (2010) Clustering Time Series Data Stream – A Literature Survey, (IJCSIS)
International Journal of Computer Science and Information Security, Vol. 8, No. 1, pp. 289-294.
[4]. S. Anitha Elavarasi and Dr. J. Akilandeswari (2011) A Survey On Partition Clustering Algorithms,
International Journal of Enterprise Computing and Business Systems.
[5]. S.Vijayalaksmi and M Punithavalli (2012) A Fast Approach to Clustering Datasets using DBSCAN and
Applications (0975 – 8887) Vol 60– No.14, pp. 1-7.
[6]. Cheng-Far Tsai and Tang-Wei Huang (2012) QIDBSCAN: A Quick Density-Based Clustering Technique
idea International Symposium on Computer, Consumer and Control, pp. 638-641.
[7]. Preeti Baser and Dr. Jatinderkumar R. Saini, A Comparative Analysis of Various Clustering Techniques
used for Very Large Datasets, International Journal of Computer Science & Communication Networks,Vol
3(4),271-275
[8]. Sunita Jahirabadkar and Parag Kulkarni (2013) Clustering for High Dimensional Data: Density based
Subspace Clustering Algorithms, International Journal of Computer Applications(0975 – 8887) Vol 63– No.20,
pp. 29-35.
[9]. Mihika Shah and Sindhu Nair, A Survey of Data Mining Clustering Algorithms, International Journal of
Computer Applications (0975 – 8887) Volume 128 – No.1, October 2015
[10]. Pavel Berkhin, Survey of Clustering Data Mining Techniques, Accrue Software, Inc.
[11]. NavneetKaur, Survey Paper on Clustering Techniques, International Journal of Science, Engineering and
Technology Research (IJSETR) Volume 2, Issue 4, April 2013.
[12]. Nisha and Puneet Jai Kaur, “A Survey of Clustering Techniques and Algorithms”, IEEE (978-9-3805-4415-
1), 2015
[13]. Han, J. and Kamber, M. Data Mining- Concepts and Techniques, 3rd Edition, 2012, Morgan Kauffman
Publishers.
[14]. Megha Mandloi, A Survey on Clustering Algorithms and K-Means, July-2014
[15]. Amandeep Kaur Mann and Navneet Kaur, “Review Paper on Clustering Techniques”, Global Journal of
Computer Science and Technology, Software and Data Engineering (0975-4350), Volume 13 Issue 5 Version
1.0 Year 2013
[16]. Pavel Berkhin, “Survey of Clustering Data Mining Techniques”, Accrue Software, Inc.

A Comparative Study Of Various Clustering Algorithms In Data Mining

Recommended

More Related Content

Similar to A Comparative Study Of Various Clustering Algorithms In Data Mining (20)

More from Natasha Grant (20)

Recently uploaded (20)

A Comparative Study Of Various Clustering Algorithms In Data Mining