A Survey of Data Clustering Methods

Abstract

Data clustering is one of the most essential, common and interesting task to classification of patterns in different areas such as data mining, pattern recognition, artificial intelligence and etc. The objective of data clustering is to classification of similar entities. There are so many different techniques of data clustering available for different nature of applications. Data clustering techniques are categorizing into two types-Partitioning Procedures and Hierarchical Procedures. Hierarchical clustering creates hierarchy of clusters, look like tree. Results of hierarchical Clusters are shown in dendrogram shape. Partitioning method-clustering makes various partitions of objects and evaluates them by some standard. In this paper, we introduce a critical review on few papers and found some strengths and weaknesses of different clustering techniques. The purpose of this overview is to compare and evaluate each clustering techniques and find their pros and cons. This comparison concludes the better approach for future research in data clustering.

Full text

International Journal of Advanced Science and Technology
Vol.113 (2018), pp.133-142
http://dx.doi.org/10.14257/ijast.2018.113.14
ISSN: 2005-4238 IJAST
Copyright ⓒ 2018 SERSC Australia
A Survey of Data Clustering Methods
1
Saima Bano and M. N. A. Khan
Shaheed Zulfikar Ali Bhutto Institute of Science and Technology,
Islamabad, Pakistan
sayohunzai@gmail.com, mnak2010@gmail.com
Abstract
Data clustering is one of the most essential, common and interesting task to
classification of patterns in different areas such as data mining, pattern recognition,
artificial intelligence and etc. The objective of data clustering is to classification of
similar entities. There are so many different techniques of data clustering available for
different nature of applications. Data clustering techniques are categorizing into two
types – Partitioning Procedures and Hierarchical Procedures. Hierarchical clustering
creates hierarchy of clusters, look like tree. Results of hierarchical Clusters are shown in
dendrogram shape. Partitioning method-clustering makes various partitions of objects
and evaluates them by some standard. In this paper, we introduce a critical review on few
papers and found some strengths and weaknesses of different clustering techniques. The
purpose of this overview is to compare and evaluate each clustering techniques and find
their pros and cons. This comparison concludes the better approach for future research in
data clustering.
Keywords: Data Clustering, Fuzzy Clustering, Fuzzy C-Means Clustering, K-Means,
K-Mediods
1. Introduction
Data clustering is a procedure in which we make cluster of entities ton based on
their similar features. A good clustering technique will create high quality clusters
with high intra-class similarity low inter-class similarity Quality of clustering
depends on the similarity measure used and its implementation. Quality of a
clustering process is measured by its ability to find out some or all of the unknown
patterns. To creating a similarity clusters distance measured is used, which are as
follow: Euclidean distance, Manhattan or taxicab distance, Mahalanobis distance,
Inner product space, Hamming distance, Cosine Similarity Index, Minkowski
distance. All distances are used to finds similarity in between different points but
mostly Euclidian distance is used to measure objects for similarity attributes. Data
Clustering is used in many different application areas; i.e., Data retrieval, Image
analysis, Machine learning, web search engines, Pattern recognition, computational,
economic, Libraries, insurances, city planning, and earthquakes studies.
There are so many data clustering algorithms to classified data into similar groups.
Normally clustering algorithms are categorized into two groups as unsupervised linear
clustering algorithm and unsupervised non-linear clustering algorithm. In unsupervised
linear clustering algorithms includes: Fuzzy c mean clustering algorithm, Quality
threshold clustering algorithm, K- mean clustering algorithm, hierarchical clustering
algorithm, and Gaussian (EM) clustering algorithm. Whereas MST created clustering
algorithm are density based and kernel k- mean are clustering algorithm include in
unsupervised nonlinear.
Received (December 27, 2017), Review Result (March 11, 2018), Accepted (March 13, 2018)

International Journal of Advanced Science and Technology
Vol.113 (2018)
134 Copyright ⓒ 2018 SERSC Australia
Some well-know clustering algorithms are K-means clustering which create a clusters
n number of objects into k clusters. Moreover, observed the nearest means to making
clusters. Assigning objects to cluster by using distance. k-means strength is fast, cover
local optimum and very flexible clusters during process its changes. K-means limitation is
when data size greater the clusters results poor because k-means always cover local
optimum clusters. While k-mediods clustering algorithms are like k-means algorithm
which n objects point into k clusters, it minimizes the dissimilarities so it is more robust
than k-means. Mediod is center point in k-mediod. It takes reference points instead of
mean values of objects.
Common clustering problems are Interpreting results, Outlier handling, Number of
clusters, Dynamic data and evaluating results, to reduce such type of issues different
clustering Procedures are used. Clustering Procedures are been classified into the
following categories:
 Hierarchical Procedures
o Agglomerative hierarchical clustering
o Divisive hierarchical clustering
 Partitioning Procedures
 Density-based Procedures
 Model-based Clustering Procedures
o Decision Trees
o Neural Networks
 Grid-based Procedures
 Fuzzy Clustering
1.1. Hierarchical Clustering Method
This method create a group of nested clusters structured as a visualized and
hierarchical tree as a dendrogram – a diagram like tree that records the structures of joins
or separations. Hierarchical procedures can be any agglomerative or divisive.
Agglomerative algorithm starts with each element as a single cluster and joins them in
sequence larger clusters; divisive algorithms start with the entire group and proceed to
split it into successively minor clusters.
1.2. Partitioning Clustering Method
The partitioning Procedures commonly result in a group of M clusters, each item
belonging to unique cluster. Each cluster may be denoted by a centroid or a cluster
representative; this is some sort of summary description of all the entities enclosed in a
cluster. The exact form of this report will depend on the nature of the entity that is being
clustered.
1.3. Density-based Clustering Method
The density –based clustering algorithm discovers the clusters in arbitrary shape. By
region of low density objects are separate into dense region. points are connecting on the
base of certain distance threshold. Points are connected till satisfy density criteria. It
handles noise and one scan. Several algorithm related to density –based is: DBSCAN,
OPTICS, DENCLUE, CLIQUE, BIRCH, CURE.
1.4. Model-based Clustering Methods
Model-based methods are used to optimize the data and some mathematical model that
is based on the assumption. Model-based clustering, model generated a data that assumes

International Journal of Advanced Science and Technology
Vol.113 (2018)
Copyright ⓒ 2018 SERSC Australia 135
and recover the original model from the data. Then clusters are defined from the
recovered data.
1.5. Grid-based Clustering Method
This method concerned with value space instead of data points to making grid
clustering. In such way grid clustering method first creates a grid structure then calculates
cell density and after that identifies cluster centers. The main advantage of grid-based
clustering is reducing a computational complexity, particularly for clustering very large
data sets. Several interesting methods of grid-based clustering method are: Wave Cluster,
STING, CLIQUE.
1.6. Fuzzy Clustering
Fuzzy clustering is referred to soft computing and the data points belong to more than
one cluster. A famous fuzzy clustering algorithm is fuzzy c- means clusters it are mostly
used in image processing tools.
2. Related Work
Karaboga et al., [1] discuses Artificial Bee Colony (ABC) algorithm an
optimizing procedure that simulates the quick seeking behaviors of a honeybee
swarm for data classification and clustering. The artificial colony consists of three
types of bees: employed bees, onlookers and scouts. Employed bee: associate with a
specific food source and offers the neighborhoods of the basis in its memory.
Onlookers: it’s get information of food source from the employed bees in the hive
and select one of the food source together the nectars. Scouts: it is responsible for
finding new food, the nectar source. The total numbers of employed bees are equal
to the number of food sources around the hive. In this paper, clustering problem is
stated i.e., N number of objects are provided and the aim of the proposed technique
is to allocate each object into k-clusters followed by minimizing the sum of
distances between the objects. ABC algorithm is compared with other famous
heuristic algorithm such as GA and PSO based on their performance. Thirteen types
of classification issues (such as glass, thyroid, and wine) from Glass Identification
Data Set, Thyroid Disease Data Set, and Wine Data Set available in the UCI
Machine Learning repository are used to evaluate the performance of the ABC
algorithm. First 75% of data used for training data, and the remaining 25% is used
for testing data. The performance is evaluated and tested using XOR, Decoder-
Encoder and 3-bit parity Procedures. The authors tested the proposed algorithm to
form clusters for patterns by assigning it to the class whose center is closest to the
cluster center. For this purpose, they have used the Euclidean distances measure.
The ABC algorithm shaped average clusters for all the problems with error
percentages of 13.13% as compared to 15.99% error percentage for PSO. ABC
algorithm offers much better quality clusters and shows good performance. The
overall ranking of the ABC algorithm was first among the other optimization
techniques.
Chen et al., [2] disuses Spectral parallel Procedures which is widely used for
computer vision and information retrieval. Author compare two types of approaches:
sparsifying the similarity matrix and the Nystro¨m approximation to discovering a
parallel spectral cluster in distributed environment. Spectral refer to the use of
eigenvalues, eigenvectors, singular values and singular vectors. It is more effective
in finding clusters than some traditional algorithms, such as k-means. The algorithm
constructs a parallel matrix and reflects the relationship between the data points,
then uses similarity matrix information to groups into k-clusters. To constructing the
sparse similarity matrix using nearest neighbor the authors use compute distances of

International Journal of Advanced Science and Technology
Vol.113 (2018)
136 Copyright ⓒ 2018 SERSC Australia
all data points, symmetrically modify the sparse matrix and finally computes the
similarities. These three steps are implemented by using MapReduce, a Google
parallel computing framework. To reduce the memory use the sparsification
approach keeps the most useful sparse matrix. Whereas, Nystrom approximation
approach used to stores only several columns of the similarity matrix. For
experimentations, three data sets were used: Corel (images), RCV1 (documents) and
Picasa-Web (a Google online platform photo sharing product. By k-means clusters
were generated an observed that spectral clustering finds better similarities in
images. By evaluation parallel spectral clustering approach speed up to 256
machines, and efficiently handle larger problem.
According to Senthilnath et al., [3] Firefly Algorithm (FA) is best for
optimization problem in clustering that is a latest nature inspired optimization
algorithm which simulates the flash pattern and characteristics of fireflies. FA is
used for difficult optimization problem in clustering. The algorithm works in
following three steps. The agents are randomly distributed in search space. The
objects were separate into classes which goal is to find clusters center. FA is
unsupervised technique so the dataset is distributed into training dataset and test
dataset. Thirteen types of data sets (The Balance data set, Cancer-Int data set and
etc.,) from UCI machine learning repository were used to shown the results of the
optimization techniques. The performance is measured by using Classification Error
Percentage (CEP) with all the 13-benchmark data sets. CEP helps to evaluate which
method has generated the optimal cluster centers. The performance of FA is
compared with other two well-known optimization algorithms i.e. Artificial Bee
Colony (ABC), Particle Swarm Optimization (PSO). The accuracy and robustness of
FA could be efficiently used for clustering and FA has good global performance
than other optimization techniques. The performance of clustering generally
depends on the size and value of training data set. FA efficiently generate clusters
center.
Kim et al., [4] presented Density-based clustering algorithms such as DBSCAN
and OPTICS are widely used in clustering. Density-based clustering Procedures are
used to discover clusters of arbitrary shape and dense regions of data points. On the
base of given density parameters, discover clusters which are dense in region. For a
large dataset author proposed a density-based clustering algorithm, which discovers
densities of cluster and well suited for framework .by using MapReduce framework
it is difficult to parallelize clustering. When dealing with large amount of data, it is
hard to parallelize clustering algorithm by using MapReduce framework. A Density-
based clustering algorithm, DBCURE find clusters with varying densities and is
suitable for parallelizing the algorithm with MapReduce. To parallelized DBCURE
using MapReduce, the authors have developed DBCURE-MR which finds several
clusters together by expanding every core point in parallel. Whereas, traditional
density-based algorithms find each cluster individually. Three datasets CLOVER,
WINDOW and BUTTERFLY have been used by the authors to evaluate clustering.
The overall findings of the study are that DBCURE and DBCURE-MR finds clusters
centers efficiently and scales up well with the MapReduce framework.
Kaymak et al., [5] used Fuzzy clustering algorithm to divides the dataset into
groups so that the clusters describe a structure within the data. Similar to fuzzy
logic, in fuzzy clustering every point has certain degree of belonging to different
clusters, rather than belonging to just one cluster. Fuzzy clustering is widely used in
various fields like finance and marketing. However, there are certain issues in fuzzy
algorithm that need to be taken care of such as volume and shape of the clusters,
distribution of the data patterns, initialization of the clustering algorithm and
selecting the number of clusters in the data.

International Journal of Advanced Science and Technology
Vol.113 (2018)
Copyright ⓒ 2018 SERSC Australia 137
Marghescu et al., [6] explore advanced supporting tools for predicting currency
crises, which is based on an experimental study of the currency crisis in 23 rising
markets around the world spanned over half century. The authors built fuzzy C-
mean (FCM) model to partition data points into specific overlapping groups and
then classify data clusters into early-warning clusters (EWCs) and tranquil clusters
(TCs). FCM model is used for predicting the overall economic crisis by testing and
evaluating a large number of samples.
Niknam et al., [7] discuses K-Mean clustering which is simple and efficient
technique to create k-clusters, and covers local optimal solutions. K-mean cluster
technique is highly depend on the initial position and finds local optimal solution.
Niknam and Amiri (2010) present a new hybrid evolutionary approach based on
FAPSO (fuzzy adaptive particle swarm optimization), ACO (ant colony
optimization) and k-means algorithms called FAPSO-ACO–K which finds global
optimal and correctly centered clusters.
Kocheturov et al., [8] analyze stock markets of the USA and Sweden by studying
the dynamics of a cluster structure in financial markets followed by finding its
correlation to crisis and non-crisis periods. The network examination has become a
great tool for learning financial markets in the last 15 years. The authors build a
network structures from a correlation matrix of the stocks markets, which has a
predefined number of connected components. Structure is a forest of stars with
weighted edges where every node represents a stock and weights are equal to the
measured similarity between the stocks.
Nanda et al., [9] presented a data mining methodology for classifying Indian
stocks market into unlike clusters. The clustering methodology classifies stocks on
certain investment criteria. Bombay Stock Exchange (BSE) was together from
Capitaline Databases Plus and data for 106 stocks were collected for the fiscal year
2007–2008. K-means, self-organizing maps (SOM) and Fuzzy C-means clustering
approaches were used to cluster stock market data. The authors collected a mixed
data from different sectors like BSE BANKEX, BSE Auto, BSE Pharma, BSE IT,
BSE Midcaps and BSE. By comparing the portfolio performances with the BSE
Sensex benchmark index, the results shows that K-means method turns out to be
better.
Sastry et al., [10] utilized clustering techniques for detecting difference in
product sales and also to identify and compare sales over a particular time.
Clustering is well suited to group items that seem to fall naturally together, when
there is no specified class for any new item. Authors used annual sales data of steel
products to analyze Sales Volume & Value with respect to dependent attributes like
products, customers and quantities sold. The demand for steel products is cyclical
and depends on many factors like customer profile, price, discounts and tax issues.
Authors have analyzed sales data with clustering algorithms like K-Means & EM
(expectation–maximization) that revealed many interesting patterns useful for
improving sales revenue and achieving higher sales volumes. K-Means & EM
(partition Procedures) algorithms are better suited to evaluate sales data in
comparison with density based Procedures.
Fallahpour et al., [11] in their study talk about applied clustering approach to
classify 79 selected stocks of Iran's stock market into a number of clusters. The data
collected from currency crisis in Iran’s economy that negatively influenced Iran
Stock Exchange dramatically during the period 22/09/2012 to 22/03/2013. Applied
three well know clustering Procedures namely K-medoids, K-means and X-means
were used. The techniques were evaluated by the application of Intraclass inertia,
which show the density of each clustering method. By comparison, of Intra-class
inertia was take that K-Means algorithm has a enhanced quality than K-medoids and
X-means techniques. By the use of some defined indexes namely Silhouette and

International Journal of Advanced Science and Technology
Vol.113 (2018)
138 Copyright ⓒ 2018 SERSC Australia
Davis-Bouldin, efficient number of clusters were extracted. Most desirable clusters
from five stock market result shows that k-means can create an efficient portfolios.
Stetco et al., [12] performed classification of registered Companies in London
Stock Exchange to indentify the group of similarly performing companies based on
their historical stock price record. Fuzzy clustering analysis were carried out using a
correlation-based metric to obtain a more insightful classification of the companies
into groups with fuzzy boundaries, giving realistic and detailed view of their
relationships. Analyze cluster and discovered groups in terms of the volatility of
their returns using both standard deviation and exponentially weighted moving
average. This approach has the potential to be of practical importance to
classification as it can detect fuzzy clusters of correlated stocks that have lower
inter-cluster correlation, analyze their volatility and sample potentially less risky
combination of assets.
Miguéis et al., [13] analyze lifestyle segmentation of customer using data mining
technique. A decent relationship between customers and companies is a vital
element of attractiveness and organization of loyalty relationships with customers is
a focal tactical purpose. Therefore, companies are improving service levels and
wishing to be at the top edge in order to certify a good business relationship with
customers. The authors [13] extracted information from a large transaction database
and propose a market segmentation technique for retailing based on customer’s
lifestyle. The authors used a variable clustering method to infer customers’ lifestyle.
The authors [13] identify typical shopping baskets based on products which are
more repeatedly bought together. Lifestyle segment is assigned to customers based
on their purchasing history. The propose model is implemented in European
retailing company.
Shim et al., [14] develop CRM approach using association rules and sequential
patterns for a small size online shopping mall. In 2002, dot-com bubble burst was
established and contained several small-sized online shopping centers. Many of
them become known in the market because they have good relations with customers
and decent characteristics of online marketplace including significantly reduced
menu cost for products/services and search costs and easily access services/products
in the world. But some of the online shopping malls have not constantly succeeded
and need to close them. Several of them have poor customer relationship
management strategies and need to close them. The authors [14] analyze customer
transactions data of the online shopping mall and propose sequential patterns and
association rules. Firstly, authors [14] define the VIP customers based on recency,
frequency and monetary (RFM) values. Formerly, they design a model which
categories customers into non-VIP and VIP. The authors [14] used various data
mining techniques such as logistic regression, bagging decision tree, artificial neural
network and decision tree with each of these as a base classifier. Lastly, they
identify patterns and rules for VIPs customer from the transactional data using
association rules and sequential patterns and then design CRM strategies for the
online shopping mall. Reference [16-25] outlined various software engineering and
machine learning techniques in different domains.
3. Critical Evaluation
Comparative evaluation of different clustering techniques as discussed above is
shown in Table 1.

International Journal of Advanced Science and Technology
Vol.113 (2018)
Copyright ⓒ 2018 SERSC Australia 139
Table 1. Critical Evaluation of Clustering Methods
Ref
#
Proposed
technique
Strengths
Limitations
Possible
Improvement
[1]
Artificial Bee
Colony (ABC)
algorithm
ABC algorithm, fast, robust,
enhances accuracy and used
for optimizing multivariable
functions.
Local search performance
depends on neighborhood
search and greedy selection
and the global search
performance of
algorithm depends on
random search process.
To establish true
strengths of
ABC algorithm,
it would be
better to
compare it with
some other
optimization
techniques such
as Genetic
Algorithms and
Ant Colony
Optimization.
[2]
Firefly
Algorithm
(FA)
The accuracy and robustness
of FA is efficient for
clustering center and
performance is better than
other optimization
techniques.
FA depends on population
size and This will affect the
computation time of this
algorithm.
To get better
results FA would
be compared
with some other
optimization
techniques like
ant colony
optimization,
genetic
algorithms etc.
[3]
Parallel
clustering
algorithm
called
DBCURE-MR.
The performance of
DBCURE-MR speedup with
an order of magnitude. And
it is more effective for
experimenting with real-life
datasets.
Time complexity is
maximum.
Because lack of
availability of
large memory,
the buffer size is
only kept to
500MB in every
machine for
simulation. To
get a better
result, the
memory size on
each machine be
enlarged.
[4]
Fuzzy c-means
(E-FCM)
extension and –
Kessel (E-GK)
algorithms
Fuzzy C-Mean
algorithm is to reduce the
sensitivity of the resulting
clustering
Algorithm miss-
classification of patterns due
to unsupervised learning.
Real world
applications of
extended
clustering
algorithms need
to be explores to
truly analyze
properties of the
algorithms.
[5]
Fuzzy C- mean
(FCM) model
FCM model is used for
predicting the overall
economic crisis by testing
and evaluating a large
number of samples.
The high rate of false alarms
in the test sets is a major
weakness of this model.
Currently, the
model is tested
for only three
Asian countries
— Indonesia,
Korea and
Taiwan. To get a
better estimate
of the results,
data about
currency crisis in
different
countries from
other continents

International Journal of Advanced Science and Technology
Vol.113 (2018)
140 Copyright ⓒ 2018 SERSC Australia
could be
included.
[6]
FAPSO (fuzzy
adaptive
particle swarm
optimization),
ACO (ant
colony
optimization)
and k-means
algorithms
called FAPSO-
ACO–K
To find a better cluster
partition and solve nonlinear
partitioned clustering
problem.
The algorithm still finds
hard optimization problem.
To get better
understanding
about its
potential use, the
proposed
algorithms could
be compared
with other
evolutionary
algorithm such
as artificial bee
colony and
firefly algorithm
etc.
[7]
Dynamic
cluster
structures
Dynamic cluster structures
proved to be more stable
during the crisis periods like
the world financial crisis, the
Subprime mortgage crisis,
the Dot-com crisis and the
banking crisis in Sweden
The modularity-based
approach reveals that the
highest values of modularity
and relatively small number
of communities help detect
the abrupt changes in the
markets such as the Dot-com
crisis is related to both types
of the markets. The world
financial crisis is not
observable from these trends
so well — as there are only
local quality threshold (Qt)
maxima and minima of the
community number. Thus
the main difference between
the modularity and our
measures is that the former
gives more contrasting
picture of the crises.
Traditionally, dynamic
clustering algorithms are
calculated for large datasets.
It is difficult to cluster the
tiny dataset due of the loss
of the statistical
characteristics and
probability features.
Author studied
the Swedish
financial market
with 266
companies. To
overcome the
world financial
crisis PMP
should be
applied more
companies and
should test the
market
performance
with other
countries.
[8]
Clustering-
based stock
selection
method
The proposed cluster-based
approach considerably
reduces the time required for
creating a portfolio. Validity
indexes were used in each
case to find the optimal
number of clusters.
Clustering algorithms were
performing on limited
dataset. The validity indexes
were no reliable in some
cases.
To get more
optimal number
of clusters in
stock market
data, it would be
better to analysis
it with some
other clustering
techniques.
[9]
K-Means and
EM
(expectation–
maximization)
Reduces the risk of
redundancy errors and
allows global integration of
product sales.
Systems are restrictive and
not flexible in
implementation and usage.
ERPs are difficult to adjust
to the specific workflow and
the main causes of their
failure. Systems can be
difficult to use.
To get better
analysis of sale
data and
resultant revenue
collection needs
to be tested on
other software
such as SAP.
[10]
K-medoids, K-
means and X-
means
Markowitz model creates
optimize portfolios on
applied best techniques and
k-means meets the portfolios
By applied method, k-means
is not guaranteed to be
global optimum.
Markowitz model is based
Other well-
known
clustering
Procedures like

International Journal of Advanced Science and Technology
Vol.113 (2018)
Copyright ⓒ 2018 SERSC Australia 141
risk minimization by
portfolio diversification.
on diversification; when
portfolios are not well
diversified then k-means is
not able to return efficient
stock portfolios.
fuzzy clustering,
expectation–
maximization
etc. can be used
to study how to
minimize
currency crisis in
the stock
exchange.
[11]
Principal
Component
Analysis
(PCA),
compression
and fuzzy
clustering
The authors computed 43
eigenvectors with 90% of the
variability in stock data. By
the using PCA, it removes
noise and to improves data
analysis performance.
1. If the time scale is kept
small (such as hourly, daily)
then not only does it
becomes computationally
more costly to analyze but
also global trends might be
difficult to discover. In
contrast, larger time
intervals may erase local
differences that may be
useful for classification.
Stock price is not a good
indicator of performance, as
it does not reflect the size
and revenue of a company;
neither is it a good
comparator when
considering multiple stocks.
2.In Fuzzy clustering, data
points have more than single
clusters so correlation cannot
be used directly as a metric
in cluster analysis as it does
not satisfy the non-negativity
condition of metric
functions.
To identify
better stock
prices for long
period, the
model-based
techniques can
be used.
4. Conclusion and Future Work
A review of different clustering techniques proposed in the literature shows that
each technique has its own advantages and disadvantages. I would like to propose a
new hybrid technique like evolutionary techniques, optimization methods such as
Genetic Algorithms and Ant Colony Optimization to compare with other techniques
to get better result of data clustering. Real world applications of extended clustering
algorithms need to be explored to analyze properties of the algorithms. The
available existing techniques regarding data clustering will overcome the issues
regarding performance, scalability and dimensionality. My focus will be to improve
data clustering technique to incorporate performance, scalability issues. The key
conclusion of this paper is to assessments of data clustering techniques, which are
used in data mining, pattern. Reason behind the review is to implement a fresh
hybrid technique in an operational method keeping in view of different data
clustering strengths and limitations based on their practicality and productivity of
current techniques are also analyzed critically.
References
[1] D. Karaboga and C. Ozturk, “A novel clustering approach: Artificial Bee Colony (ABC) algorithm”,
Applied Soft Computing, vol. 11, no. 1, (2011), pp. 652-657.
[2] W. Y. Chen, Y. Song, H. Bai, C. J. Lin and E. Y. Chang, “Parallel spectral clustering in distributed
systems”, Pattern Analysis and Machine Intelligence, IEEE Transactions, vol. 33 no. 3, (2011), pp. 568-
586.

International Journal of Advanced Science and Technology
Vol.113 (2018)
142 Copyright ⓒ 2018 SERSC Australia
[3] J. Senthilnath, S. N. Omkar and V. Mani, “Clustering using firefly algorithm: performance study”,
Swarm and Evolutionary Computation, vol. 1, no. 3, (2011), pp. 164-171.
[4] Y. Kim, K. Shim, M. S. Kim and S. Lee, “DBCURE-MR: an efficient density-based clustering algorithm
for large data using MapReduce”, Information Systems, vol. 42, (2014), pp. 15-35.
[5] U. Kaymak and M. Setnes, “Extended fuzzy clustering algorithms”, ERIM Report Series Reference No.
ERS-2001-51-LIS, (2000).
[6] D. Marghescu, P. Sarlin and S. Liu, “Early-warning analysis for currency crises in emerging markets: A
revisit with fuzzy clustering”, Intelligent Systems in Accounting, Finance and Management, vol. 17, no.
3-4, (2010), pp. 143-165.
[7] T. Niknam and B. Amiri, “An efficient hybrid approach based on PSO, ACO and k-means for cluster
analysis”, Applied Soft Computing, vol. 10, no. 1, (2010), pp. 183-197.
[8] A. Kocheturov, M. Batsyn and P. M. Pardalos, “Dynamics of cluster structures in a financial market
network”, Physica A: Statistical Mechanics and its Applications, vol. 413, (2014), pp. 523-533.
[9] S. R. Nanda, B. Mahanty and M. K. Tiwari, “Clustering Indian stock market data for portfolio
management”, Expert Systems with Applications, vol. 37, no. 12, (2010), pp. 8793-8798.
[10] S. H. Sastry, P. Babu and M. S. Prasada, “Analysis & Prediction of Sales Data in SA P-ERP System
using Clustering Algorithms”, arXiv preprint arXiv:1312.2678, (2013).
[11] S. Fallahpour, M. H. Zadeh and E. N. Lakvan, “Use of Clustering Approach for Portfolio Management”,
International SAMANM Journal of Finance and Accounting, vol. 2, no. 1, (2014).
[12] A. Stetco, X. Zeng and J. Keane, “Fuzzy cluster analysis of financial time series and their volatility
assessment”, In Systems, Man, and Cybernetics (SMC), IEEE International Conference, (2013), pp. 91-
96.
[13] V. L. Miguéis, A. S. Camanho and J. F. e Cunha, “Customer data mining for lifestyle segmentation”,
Expert Systems with Applications, vol. 39, no. 10, (2012), pp. 9359-9366.
[14] B. Shim, K. Choi and Y. Suh, “CRM strategies for a small-sized online shopping mall based on
association rules and sequential patterns”, Expert Systems with Applications, vol. 39, no. 9, (2012), pp.
7736-7742.
[15] C. Pete, C. Julian, K. Randy, K. Thomas, R. Thomas, S. Colin and W. Rüdiger, “CRISP-DM”, NCR,
SPSS, DaimlerChrysler. USA, (2000), pp. 1-76.
[16] M. N. A. Khan and S. Ullah, “A log aggregation forensic analysis framework for cloud computing
environments”, Computer Fraud & Security, vol. 2017, no. 7, (2017) July, pp. 11-16.
[17] M. N. A. Khan and I. Wakeman, “Machine learning for post-event timeline reconstruction”, In First
Conference on Advances in Computer Security and Forensics Liverpool, UK, (2006), pp. 112-121.
[18] S. Rahman and M. N. A. Khan, “Review of live forensic analysis techniques”, International Journal of
Hybrid Information Technology, vol. 8, no. 2, (2015), pp. 379-88.
[19] M. N. A. Khan, C. R. Chatwin and R. C. Young, “Extracting Evidence from Filesystem Activity using
Bayesian Networks”, International journal of Forensic computer science, vol. 1, (2007), pp. 50-63.
[20] M. N. A. Khan, “Performance analysis of Bayesian networks and neural networks in classification of file
system activities”, Computers & Security, vol. 31, no. 4, (2012), pp. 391-401.
[21] M. N. A. Khan, C. R. Chatwin and R. C. Young, “A framework for post-event timeline reconstruction
using neural networks”, Digital Investigation, vol. 4, no. 3-4, (2007), pp. 146-157.
[22] M. S. Bashir and M. N. A. Khan, “Triage in live digital forensic analysis”, International journal of
Forensic Computer Science, vol. 1, (2013), pp. 35-44.
[23] M. Rafique and M. N. A. Khan, “Exploring static and live digital forensics: Methods, practices and
tools”, International Journal of Scientific & Engineering Research, vol. 4, no. 10, (2013), pp. 1048-1056.
[24] R. Shehzad and M. N. A. Khan, “Integrating knowledge management with business intelligence
processes for enhanced organizational learning”, International Journal of Software Engineering and Its
Applications, vol. 7, no. 2, (2013), pp. 83-91.
[25] M. Khalid, S. ul Haq and M. N. A. Khan, “An assessment of extreme programming based requirement
engineering process”, International Journal of Modern Education and Computer Science, vol. 5, no. 2,
(2013), pp. 41.