IRJET- Comparatively Analysis on K-Means++ and Mini Batch K-Means Clustering Algorithm in Cloud Computing with Map Reduce

International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 06 Issue: 09 | Sep 2019 www.irjet.net p-ISSN: 2395-0072
© 2019, IRJET | Impact Factor value: 7.34 | ISO 9001:2008 Certified Journal | Page 814
Comparatively Analysis on K-Means++ and Mini Batch K-Means
Clustering Algorithm in Cloud Computing with Map Reduce
Virendra Tiwari1, Dr.Akhilesh A. Waoo2
1
Assistant Professor, Dept. of Computer Science & Application, AKS University, Satna, Madhya Pradesh, India
2
Associate Professor, Dept. of Computer Science & Application, AKS University, Satna, Madhya Pradesh, India
-----------------------------------------------------------------------***--------------------------------------------------------------------
Abstract - Cloud computing provides the functionality to host
the servers in a dispersed network so wide range of customers
can access the applications asutilities over the internet. Oneof
the very important aspects for the development of cloud
computing is the rapidly growing the volume ofBigdatawhich
requires to be managed and examined efficiently. Experts and
Researchers design and utilize multiple algorithms with
intelligent approach to gain useful knowledge from a huge
volume of data set. Hadoop MapReduce is a new and emerging
software platform suitable for writing applications easily
which process huge amounts of data in-parallel on large
clusters of commodityhardwareinareliableandfault-tolerant
manner. There only specific unsupervised learning algorithms
for big data problems which are working and being executed
successfully in MapReduce technique and are deployedonhigh
volume of data set. MapReduce frameworks performance
suffers degradation due to the sequential processing
approaches. To overcomethislimitationandreducingcosts,we
can introduce cloud computing with parallel processing in
MapReduce a powerful and effective approach for the future
technology enhancements. This survey givesabriefoverviewof
cloud computing and comparatively analysis on most popular
clustering algorithms named K-means++ and Mini Batch K-
means.
Key Words: Cloud computing, Big data, Hadoop,
MapReduce, K- Means++, and Mini Batch K-means.
1. INTRODUCTION
Cloud computing in simple terms on demand availability of
hardware resources for computing or storage purpose over
the Internet instead of ourcomputer'sharddrive. Withcloud
computing, clients can access files and many applications
from any remote device through internet. It reduces the
work and cost for client to setup and manage the hardware
and required infrastructure [11][12]. Cloud provider take
the ownership of providing the set up and maintaining the
hardware infrastructure. Most famous cloud service
providers are Amazon, Alibaba, Google and Microsoft [13].
Most famous Cloud Service Providers [11] [14]
a) Google: Google Cloud Platform solves issues with
accessible AI & data analytics and uses resources
such as hard disks, computers,virtual machines,etc.
located at Google data centers. Googleisa dedicated
cloud computing service provider, with all the
storage available online so it can work with the
cloud apps: Google Sheets, Google Docs, andGoogle
Slides. Most of Google's services could be
considered cloud computing: Gmail, Google Maps,
Google Calendar, Picasa, Google Analyticsandsoon.
b) Alibaba: Alibaba is one the largestcloudcomputing
platform which created a global footprint with over
1500 CDN Nodes worldwide of 19 regions and 56
availability zones across more than approximate
200 countries. Its prominent features arehelpingto
protect and backup your data and help you to
achieve faster results.
c) Apple iCloud: Apple's cloud service is mainly used
for online backup, storage, and synchronization of
your mail, calendar, contacts, and more. All thedata
we need remains available on Mac OS, iOS, or
Windows device.
d) Amazon Cloud Drive: Web hosting platformwhich
storage reliable and cost-effective solutions. If you
have Amazon Prime, you get unlimited image
storage. It's essentially storage for anything digital
and It is the most popular as it was the first to enter
the cloud computing space.
World has created 90% of the data in the last two years
which comes from everywhere and surely this is only going
to grow with arrival of internet of things. By 2020, it’s
estimated that approx 1.7MB of data will be created every
second for every person on planet"[13]. Processing such a
huge amount of data will be extreme challengingforthedata
analyzers and researchers. Classical sequential methods of
programming are not very efficient and hence the
requirement of parallel processinggainedimportance.Cloud
provides the most suitable environment and Hadoop
MapReduce as the processing components a programming
paradigm that enables massive scalability across hundreds
or thousands of servers in a Hadoop cluster [16][17].
Virtually we are floating in oceans of data, with theadvent of
Big data; MapReduce emerged as a most effective solution.
Number of researchers work is undergoing on the extension
of MapReduce carried out with new features, functionalities
and mechanism to perfect it for a new set of problems
[18].Various sequential programming algorithms are
nowadays being use to convert to Hadoop MapReduce
programming paradigm. There may be few algorithms that
cannot be parallelized, so their uses due to working
efficiency are less in the real world as parallelization is of
essence to process Big Data. Hence some of the conventional
useful algorithms for clustering may provide a great

platform and enhancements tothenewemergingtechnology
and also very beneficial to the business enterprise
organizations [13].
2. BACKGROUND
a) Cloud computing
Cloud computing is the delivery of different services like
managing hardware and software through the Internet. It
has evolved since the inception of internet due to its
efficiency in storing data, computationandlessmaintenance
cost. Cloud-based storage is making it possible to save files
and documents to a remote database and retrieve them on
demand. Services can be in public and private. Public
services are providing online for a fee while private services
are being hosted on a network to particular clients [19].
Figure. 1: Cloud computing services type
Figure. 2: Cloud computing deployment models
As shown in Fig. 1, cloud providers services can be
categorized in three delivery types that are Software as a
Service (SaaS), Infrastructure as a Service (IaaS) and
Platform as a Service (PaaS). Cloud deployment services
comes in the form of private, hybrid, public and community
based on their location as mentioned in Figure 2.Public
clouds based on a shared cost model have many customers
compared to other forms of cloud. Private clouds are perfect
for organizations that have high management demands,
high-security requirements, and availability requirements
usually but are not cost efficient. CommunityCloudmutually
shared model between organizations [21].
b) Hadoop
Hadoop is an open-source software framework considered
to be one of the best tools to handle and analyze very huge
volume of data. Currently being used by Google, Facebook,
LinkedIn, Yahoo, Twitter etc. It has two major components
Hadoop Distributed File System (HDFS) and another is
MapReduce. Hadoop is being highly in used distributed
Computing platform specially in past 10 years as big data is
getting evolved with social media generating PETA bytes of
data everyday which can be used for data mining, predictive
analytics, and machine learning applications. Hadoop can
run with single node or multi-node cluster designed
specifically for storing and analyzing huge amounts of
unstructured data in a distributed computing environment.
When HDFS takes data as input, it breaks the information
down into various separate blocks and distributes all the
sepreted blocks to different nodes in a cluster; it provides
highly efficient parallel processing. HDFS can be thought of
Data node (stores actual data in HDFS) + Name node( holds
the meta data for the HDFS ) + Secondary Name
node(merging editlogs with fsimage from the namenode)
and daemon process to manage MapReduce programming
paradigm in HDFS are Job Tracker[20][13].
Figure. 3: Data flow in HDFS
c) Map reduce paradigm
Map Reduce Programming Paradigm technique and a
program model for distributed computing to process huge
amount of data. MapReduce program executes in three
stages: Map phase process and maps the input and creates
several small chunks of data into key-value pairs
[3][22].Complete process gets executed by four phases
namely splitting, mapping, shuffling, and reducing. The
reduce phase combines the data based on common keys and
performs reduce operation defined by the user. The
parallelization implementation occurs with many mappers
created for reading the data and it is not sequential. Because
of this there is high throughput [5][21].The core advantage

of the MapReduce framework is its fault tolerance due to
periodic reports from each node in the cluster are expected
when work is processed.
Figure. 4: Map reduce in HDFS
3. K-Means++ CLUSTERING ALGORITHM
K-Means++ algorithm is the improved and robust version of
most popular and widely used K-means algorithm in
clustering. K-Means algorithm chooses and initializes the
centroid randomly, forms the clusters for a given dataset
[23].
The performance and accuracy of k-means depends
completely on the goodness of the initialization where this
algorithm observed and found sometimes giving less
accurate clusters. To overcome this limitation, we can use
improved clustering method K-Means++ which overcomes
by initializationparttoimproveK-Meansalgorithm[13][24].
K- Means++ technique, takes an input k, which refers to the
number of clusters that should be generated and n refers to
set of objects [4]. This improved algorithms mainly aims to
initialize the centroids for traditional k-means clustering
algorithm [25].K-Means++ algorithm reduces the
computational time and it can handle huge amount of data
sets.
K-Means++ clustering is one of the partitioning based
clustering algorithm works as follows [25].
[1] Choose initial centroid X uniformly at random from
initial data set.
[2] For each data point X we need to compute D(X)
which is the minimum squared distance between X
and the nearest centroid that has been already
defined.
[3] Choose another data point as second centroidusing
a weighted probability distribution which is
proportional to D(X).
[4] Repeat 2 and 3 until required K centroidshavebeen
chosen.
[5] Assign each record or instance point to a cluster
centre which has least distance.
[6] Replace cluster centroids by calculatingmeanvalue
of all points in the cluster.
In map-reduce pattern mentioned algorithm can be
implemented as follows.
 Map function: The HDFS stores input data as
sequence file of <key, value> pairs and produces a
set of <key, value> pairs as the output of the job[6].
The map function splits the data acrossall mappers,
creates a map task for each split and assigns each
map task to a Task Tracker[7].First Step is Map,
Second Step is Shuffle and Third Step is
Reduce[26][27].
 Reduce function: Aftermapping, reducersareused
for computing Step-2 of the mentioned algorithm.
The Reducer’s job is to process the data that comes
from the mapper [8]. After processing, the
intermediate data can be put in hdfs or stored
locally [9]. Afterwards new centers will be
generated which can be used for further iterative
statements [27].
Figure. 5: K-Means++ Map Reduce
4. Mini Batch K-means
The mini batch k-means clustering algorithm [29] is the
alternative and modified version of the k-means algorithm.
The advantage of this algorithm uses mini-batchestoreduce
the computation time and cost in large datasets not using all
the dataset each iteration [33]. Moreover, it attempts to
provide optimal result of the clustering. To accomplish this,
the mini batch k-means takes mini-batches of a fixed size as
an input randomly, which are subsets of the all dataset [30].
The Mini Batch K-Means framework with Map-Reduce
[31][32]:
1. From the throughput of the input sequence, initial cutoff
gets calculated using a sample of points.
2. Here, MapReduce job splits the input filesintochunks and
further assign each split to a mapper to process thatconsists
of an instance of the k-means algorithm this is called
InputSplit.

3. It will be giving a set of centroids c that is much larger
than the final expected clusters k but, it should be small
enough to fit into the memory.
4. Now, Each split which is the results of clustering gets pass
to single reducer uses thestreamingBall k-meansfunctionto
combine the clusters as required.
Before invoking the Map routine, OutputFormat organizes
the key-value pairs from Reducer for writing it on HDFS.
Based on optimal distance-cutoff parametersarerequiredto
measure the distance between the point and the centroids,
Either the point is merged to any of the existing clusters or
become a new centroids to form new cluster. To determine
Optimal Clusters, the points which are forming the clusters
with any of centroids become the value in key-value pair.
5. CONCLUSIONS AND FUTURE WORK
Today, Big Data and Cloud Computing are the two
mainstream technologies to managing and processing big
data Big data makes you dealing with the massive scale of
data whereas Cloud computing provides you infrastructure.
Today Big Data generating the demand of efficient
computing, well managed storage resources as well as best
way to process the data [27]. Conventional methods are not
found by the researchers and failing to cope up with such
challenges because of which Cloud Computing and Hadoop
MapReduce found efficient and gained popularity [10].
MapReduce is intelligent enough to tackle huge data sets by
distributing processing across many nodes, and then
reducing or combining the results of those nodes. In
summary, Mini Batch K-means and K-Means++ algorithms
discussed and analyzed in this research are robust and
efficient for large set of data. K-means++ always tries to
select centroids that are far away from the existing
centroids, which gives significant improvement over others
with the issue of outlier whereas Mini Batch K-means
algorithm is not only found suitable and efficient for the
processing of huge data, but also make sure the accurate by
reduce the amount of computation required [28][31]. Italso
found giving relatively stable precision, for the outlier
detection of data. For scaling well and efficiently process
large datasets this algorithm requires to be improved in a
way so it could reduce outliers efficiently and to give better
and more accurate results with reduced run time.
REFERENCES
[1]Rajashree Shettar, Bhimasen. V. Purohit, “A Review on
Clustering Algorithms Applicable for Map Reduce”,
International Conference on Computational Systems for
Health & Sustainability, pp. 176-178, 2015.
[2]Adem Tepe, GüRay Yilmaz,“ASurveyonCloudComputing
Technology and Its Application to SatelliteGroundSystems”,
International Conference on Recent Advances in Space
Technologies (RAST), 2013.
[3]K. Singh and R. Kaur, "Hadoop: Addressing challenges of
Big Data," 2014 IEEE International Advance Computing
Conference (IACC), Gurgaon, 2014, pp. 686-689.
[4]Weihzong Zhao, Huifang Ma, Qing He, “Parallel K-Means
clustering Based on MapReduce”, springer-verlag Berlin,
Heidelberg, 2009.
[5]Borthakur. D “The Hadoop Distributed File System:
Architecture and design”, 2007.
[6]Sangeeta Ahuja, M.Ester, H. P. Kriegel, J. Sander, X. Xu, “A
Density based algorithm for discovering clusters in large
spatial database with noise”, Second international
conference on knowledge discovery and Data Mining,1996.
[7]B. Dai and I. Lin, "Efficient Map/Reduce-Based DBSCAN
Algorithm with Optimized Data Partition," 2012 IEEE Fifth
International Conference on Cloud Computing,Honolulu,HI,
2012, pp. 59-66.
[8]V. Gaede, O. G’unther,“Multidimensional accessmethods”,
ACM comput. Surv., Vol. 30, No. 2, pp. 170- 231, 1998.
[9]VaradMeru“Data clustering:UsingMapReduce”,software
Developers Journal, 2013.
[10]Das A.S, Datar M, Garg, and Rajaram S, “Google news
personalization scalable online collaborative filtering”, pp.
271-280, 2007.
[11]Aaqib Rashid and Amit Chaturvedi, "Cloud Computing
Characteristics and Services", 2019.
[12]Mahantesh N. Birje and Praveen S. Challagidad "Cloud
computing review: concepts, technology, challenges and
security “, 2017.
[13]Sweekruth S Badiger and Sushmitha N,"A Review on K-
means++ ClusteringAlgorithminCloudComputingwith Map
Reduce”, 2019.
[14]Dr. Richa Purohit,"Comparative Analysis of Few Cloud
Service Providers Considering Their DistinctiveProperties”,
2017.
[15]Vishal R. Pancholi and Dr. Bhadresh P. Patel,"A Study on
Services Provided by Various Service Providers of Cloud
Computing", 2017.
[16]https://www.ibm.com/analytics/hadoop/mapreduce.
[17]Dr. Anitha Patil,"Securing mapreduce programming.
Paradigm in hadoop, cloud and big data eco-system", 2018.
[18]Jeffrey Dean and Sanjay Ghemawat,"MapReduce:
Simplified Data Processing on Large Clusters", 2004.
[19]Palvinder,"Survey Paper on Cloud Computing”, 2014.

[20]Nasrullah and Ms. Akanksha Bana, "Review paper on
hadoop configuration and implementation in virtual cloud
environment", 2018.
[21]Santhosh voruganti,"Map Reducea ProgrammingModel
for Cloud Computing Based On Hadoop Ecosystem", 2014.
[22]Bisma Bashir, "An Approach of MapReduce
Programming Model for Cloud Computing", 2017.
[23]Md. Zakir Hossain, Md. Nasim Akhtar, R.B. Ahmad and
Mostafijur Rahman,"A dynamic K-means clustering for data
mining ", 2019.
[24]Dhruv Sharma, Krishnaiya Thulasiraman and John N.
Jiang, "A network science-based k-means++ clustering
method for power systems network equivalence", 2019.
[25]G. Yamini and Dr. B. Renuka Devi, "A New Hybrid
Clustering Technique Based on Mini-batch K-means And K-
means++ For Analysing Big Data", 2018.
[26]N. Deshai, S. Venkataramana and Dr. G. P. Saradhi
Varma3 "Research Paper on Big Data Hadoop Map Reduce
Job Scheduling", 2018.
[27]Ch. Shobha Rani and Dr. B. Rama"MapReduce with
Hadoop for Simplified Analysis of Big Data", 2017.
[28] Bo Xiao, Zhen Wang, Qi Liu, and Xiaodong Liu "SMK-
means: An Improved Mini Batch K-means Algorithm Based
on Mapreduce with Big Data", 2018.
[29]Sculley D , "Web-scale k-means clustering", 2010.
[30]Ali Feizollah, Nor Badrul Anuar, Rosli Salleh and Fairuz
Amalina, "Comparative Study of K-means and Mini Batch K-
means Clustering Algorithms in Android Malware Detection
Using Network Traffic Analysis",2014.
[31]Meghana M Chavan, Asawari Patil,Lata Dalvi And
Ajinkya Patil, "Mini Batch K-Means Clustering On Large
Dataset",2015.
[32] Mr. Krishna Yadav and Mr. Jwalant Baria,"Mini-BatchK-
Means Clustering Using Map-Reduce in Hadoop", 2014.
[33]Nadeem Akthar, Mohd Vasim Ahamad and Shahbaaz
Ahmad, "MapReduce Model ofImprovedK-MeansClustering
Algorithm Using Hadoop MapReduce", 2016.

IRJET- Comparatively Analysis on K-Means++ and Mini Batch K-Means Clustering Algorithm in Cloud Computing with Map Reduce

More Related Content

What's hot (20)

Similar to IRJET- Comparatively Analysis on K-Means++ and Mini Batch K-Means Clustering Algorithm in Cloud Computing with Map Reduce (20)

More from IRJET Journal (20)

Recently uploaded (20)

IRJET- Comparatively Analysis on K-Means++ and Mini Batch K-Means Clustering Algorithm in Cloud Computing with Map Reduce