Pramana Research Journal ISSN NO: 2249-2976

K- Means Clustering Techniques - A Review

Arshleen
Department of CSE, Chandigarh University, Gharaun

Abstract - The Clustering is a fundamental 1. CLUSTERING

technique of categorizing data into various
groups such that the data belonging to one The process of clustering plays an important
group is similar to each other but they are very role in the analysis and mining of data in
distinct from data existing in the other groups. various applications [2]. The data is divided
K Means Clustering method involves division into distinct classes on the basis of its
of any given data set into k number of clusters. attributes and qualities.
This algorithm is also known as nearest The clustering comes under the category of
neighbor clustering. K means clustering is the Unsupervised learning in which the expected
most imperative and basic technique for the outcome is not given during method of
investigation of various data sets. Beforehand, learning. Here, the data is clustered into groups
different endeavors have been done to enhance on the basis of their statistical properties.
the execution of K-means algorithm. The Another category of clustering known as
result of enhanced k-means has given great Supervised learning involves a trainer who
performance in terms of small to medium sized gives input, and the desired output is
data in comparison to the huge and substantial generated. Many algorithms are formulated to
data. This paper reviews the various strategies accomplish clustering.
and techniques used in the literature along with
its benefits and shortcomings. The paper
would also investigate the future possibilities
of advancements in k means algorithm.
RAW INPUT
Keywords - Clustering, K-Means, Nearest DATA
Neighbor, Data points.

INTRODUCTION

With the passage of years a sharp growth is

seen in the Internet usage. This increase in the
web utilization results in producing heaps of CLUSTERING
data which is expanding as the years passes. ALGORITHM
The investigation of such data and grouping it
into clusters is a difficult job. Further, issue
lies in putting away and recovering the data.
The scientists have evaluated that the data
becomes two fold after every 20 months.

Nonetheless the crude information cannot be DATA

utilized straightforwardly. Its genuine worth is CLUSTERS
anticipated by separating data that is helpful
for making decisions. The investigation of data
was a manual procedure in many areas. But
now the individuals are searching for figuring
innovations to mechanize the procedure [1] as
the measure of information control and Figure1. Clustering Stages
investigation is going past human abilities.

There are two broad categories of clustering

Volume 8, Issue 3, 2018 97 https://pramanaresearch.org/

Pramana Research Journal ISSN NO: 2249-2976

namely hierarchical and partitioning distinctive groups. Last group quality in calculation
clustering. relies upon the choice of starting centroids. Two
stages incorporate into unique k implies
1. Hierarchical Clustering: It consists of calculation: first for deciding beginning centroids
clusters which are nested and arranged in and second to allot information focuses to the
the form of tree. closest bunches and afterward recalculating the
2. Partitioning Clustering: It consists of grouping mean.
dividing various data into subsets so that
each object of data is exactly in one Soumi Ghosh et al. [7] present a relative talk of two
subset. bunching calculations in particular centroid based
K-Means and agent question based FCM (Fluzzy
This paper reviews the literature for the C-Means) grouping calculations. This exchange is
various procedures and techniques used for K based on execution assessment of the effectiveness
Means clustering. It also discusses the benefits of bunching yield by applying these calculations.
and drawbacks of the method along with the
chances for the enhancement of K means Shafeeq et al. [8] present an altered K-implies
algorithm. calculation to enhance the group quality and to
settle the ideal number of bunch. As info number of
3. K- MEANS CLUSTERING bunches (K) given to the K-implies calculation by
the client. In any case, in the handy situation, it is
Clustering is a vital and an imperative topic in exceptionally hard to settle the number of groups
the field of data mining which is effectively ahead of time. The technique proposed in this paper
used in different applications. The process of works for both the cases i.e. for known number of
clustering involves the division of data into groups ahead of time and also obscure number of
distinct classes where each class of data has groups. The client has the adaptability either to
significant properties [3]. Therefore it can be settle the quantity of groups or information the base
concluded that classes comprise the objects number of bunches required. The new bunch
having exactly similar attributes. focuses are figured by the calculation by
augmenting the group counter by one in every
The K- Means clustering is the widely used cycle until the point when it fulfills the legitimacy
clustering tool in the various fields of scientific of bunch quality. This calculation will survive this
and mechanical applications [4]. It is a strategy issue by finding the ideal number of bunches on the
used for cluster investigation in which various run.
observations are classified into k clusters and
each of the observation belongs to the cluster Junatao Wang et al. [9] propose an enhanced k
with the nearest mean [5]. means calculation utilizing commotion information
channel in this paper. The inadequacies of the
The K-Means algorithm is simple and easy to customary k-implies bunching calculation are
understand. The various steps are: overwhelmed by this proposed calculation. The
calculation creates thickness based identification
 Select the value of k as initial techniques in light of qualities of clamor
centroids. information where the revelation and preparing
 Repeat the following two steps for all ventures of the clamor information are added to the
the points in a set of data. first calculation. By pre-handling the information to
 Form k cluster by allocating every prohibit these clamor information before bunching
point to its most nearest centroid. information sets the group attachment of the
 Recalculate the centroid for each bunching results is enhanced altogether and the
cluster until the centroid does not effect of commotion information on k means
change. calculation is diminished adequately and the
grouping results are more exact.
The algorithm is used to a great extent in the field
of data mining to extract useful data from the large Shi Na et al. [10] present the investigation of
sets of data. weaknesses of the standard k-implies calculation.
As k means calculation needs to figure the
separation between every datum protest and all
4. LITERATURE SURVEY bunch focuses in every emphasis.. An enhanced k-
implies calculation proposed in this paper.
K. A. Abdul Nazeer et al. [6] proposes k-implies
calculation, for various arrangements of
estimations of beginning centroids, produces

Volume 8, Issue 3, 2018 98 https://pramanaresearch.org/

Pramana Research Journal ISSN NO: 2249-2976

Table 1 Comparison among various existing Approaches and its Limitations

S.No. Author Methodology Used Country of Objectives Limitations

Research

1. K. A. Abdul Nazeer et K-Means Algorithm India Presented an The limitation

al improved of this algorithm
clustering method lies on the fact
that calculates the that despite the
initial centroid distribution of
values and also the various data
effectively points, it is still
allocated the data required to give
points to the count of clusters
clusters. as an input
Enhanced the
correctness of K-
Means Algorithm.

2. Soumi Ghosh et al. K-Means Algorithm, India Performs relative The calculation
Fuzzy C- Means investigation of time taken is
Algorithm Fuzzy C Means more because of
and K means the fuzzy
algorithm based on measurements.
the criteria of time
complexity.
K- Means
algorithm seems
far better than
Fuzzy C-Means.

3. Shafeeq et al. Modified India The exact number The technique

K-Means Algorithm of devised on the devised takes
basis of run more time for
method of calculation than
clustering. k means in case
It works good for of big data sets.
both familiar and
non-familiar
number of clusters.

4. Junatao Wang et al. K-Means Algorithm China The updated The noise
algorithm produces impact is more
less nose data as in cluster
compared to the forming.
earlier researches

5. Shi Na et al K-Means Algorithm China Enhances the The algorithm

speed and used for the
decreases the selection of the
calculative centroid is not
complexity. very effective.

Volume 8, Issue 3, 2018 99 https://pramanaresearch.org/

Pramana Research Journal ISSN NO: 2249-2976

5. CONCLUSIONS

In this paper k-implies grouping strategies and 224, 8-10 April 2011
strategy are checked on. K-implies being generally
acclaimed among information researcher requires [5] Amandeep Kaur Mann, Navneet Kaur Mann,
promote change in different area of calculation. Review Paper On Clustering Techniques ,Global
The exceptions, void groups what's more, choosing Journal Of Computer Science And Technology
centroid for datasets are as yet a testing errand. Software & Data Engineering, VOL. 13 ,201
Thus different further research expected to center
around these said issues. Table I. presents different [6] K. A. Abdul Nazeer, M. P. Sebastian,
methods and its restriction is available in proposed Improving the Accuracy and Efficiency of the k-
k means calculation. They require facilitate means Clustering Algorithm, Proceedings of the
improvement due to increment of size of World Congress on Engineering 2009 Vol I WCE
information starting at now. This paper has make 2009, July 1 - 3, 2009, London, U.K.
an endeavor to survey a huge number of papers to
manage the present calculation of k-implies. [7] Soumi Ghosh, Sanjay Kumar Dubey,
Present examination show that k means calculation Comparative Analysis of K-Means and Fuzzy C-
can be upgraded by choosing centroid point Means Algorithms, International Journal of
fittingly. Advanced Computer Science and Applications,
Vol. 4, No.4, 2013
REFERENCES
[8] Shafeeq, A., Hareesha ,K., Dynamic Clustering
[1] E. A. Khadem, E. F. Nezhad, M. Sharifi, “Data of Data with Modified K-Means Algorithm,
Mining: Methods & Utilities”, Researcher 2013; International Conference on Information and
5(12):47-59. (ISSN: 1553-9865). Computer Networks, vol. 27 ,2012

[2] Namrata S Gupta, Bijendra S Agrawal, [9] Junatao Wang, XiaolongSu, An Improved K-
Rajkumar M. Chauhan, Survey On Clustering means Clustering Algorithm, Communication
Technique of Data Mining, American International Software and Networks (ICCSN), 2011 IEEE 3rd
Journal of Research in Science, Technology, International Conference on 27 may,2011 (pp. 44-
Engineering & Mathematics, ISSN:2328-3491 46)

[3] Malwinder Singh, Meenaksh bansal , A Survey [10] Shi Na, Liu Xumin, Guan Yong, Research on
on Various K Means algorithms for Clustering, K-means Clustering Algorithm: An Improved K
IJCSNS International Journal of Computer Science means Clustering Algorithm, Intelligent
and Network Security, VOL.15 No.6, June 2015. Information Technology and Security Informatics,
2010 IEEE Third International Symposium on 2-4
[4] A. Saurabh, A. Naik, "Wireless sensor network April, 2010(pp. 63-67)
based adaptive landmine detection algorithm, "
2011 3rd International Conference on Electronics
Computer Technology (ICECT), vol.1, no., pp.220,

Volume 8, Issue 3, 2018 100 https://pramanaresearch.org/