06. 9534 14985-1-ed b edit dhyan

International Journal of Informatics and Communication Technology (IJ-ICT)
Vol.8, No.1, April 2019, pp. 39~49
ISSN: 2252-8776, DOI: 10.11591/ijict.v8i1.pp39-49  39
Journal homepage: http://iaescore.com/journals/index.php/IJICT
Recent trends in big data using hadoop
Chetna Kaushal, Deepika Koundal
Department of Computer Science and Engineering, Chitkara University, India.
Article Info ABSTRACT
Article history:
Received Nov 9, 2018
Revised Dec 25, 2018
Accepted Jan 11, 2019
Big data refers to huge set of data which is very common these days due to
the increase of internet utilities. Data generated from social media is a very
common example for the same. This paper depicts the summary on big data
and ways in which it has been utilized in all aspects. Data mining is radically
a mode of deriving the indispensable knowledge from extensively vast
fractions of data which is quite challenging to be interpreted by conventional
methods. The paper mainly focuses on the issues related to the clustering
techniques in big data. For the classification purpose of the big data, the
existing classification algorithms are concisely acknowledged and after that,
k-nearest neighbour algorithm is discreetly chosen among them and
described along with an example.
Keywords:
Big Data
Classification
Clustering
Knowledge Discovery
Mining Copyright © 2019 Institute of Advanced Engineering and Science.
All rights reserved.
Corresponding Author:
Chetna Kaushal
Department of Computer Science and Engineering,
Chitkara University, Punjab, India
1. INTRODUCTION
The Big data is primarily defined as a term that generally describes the large dimensions of high
velocity, difficult and variable data that involve innovative techniques and equipment to facilitate the capture,
storage, sharing, administration, and analysis of the data or information [1]. Big data ultimately surpasses the
handlingability of traditional databases and is too big to be managed by a single machine. Therefore, novel
and advanced ways are mandatory to process and store such an enormous size of the data. These data are
produced from virtual transactions, electronic mails, audios, videos, pictures, torrents, records, posts, search
requests, fitness records, social networking connections, science data, sensors and cell-phones and their
applications [2]. They are deposited in databases that rise enormously and turn out to be challenging in order
to capture, arrange, store, manage, share and analyze the database with the use of standard database software
tools.
Database Machine is an important part of Big data processing. The idea of the “database machine”
was first appeared in the late 1970’s, it is an equipment that was specially built for the purpose of storage and
analysis of data. A sole mainframe network arrangement turned insufficient with the increment of data extent
and the data repository. With the increasing demand of technology, Teradata system emerged as the leading
commercially efficient database which was based upon the parallel system. In 1986, a breakthrough event
happened while Teradata initially brought the system of parallel database comprising the capacity of storing
data from 1TB up to Kmart in order to provide convenience to the retail companies at large-scale. The
benefits of the parallel system based database stood broadly acknowledged in the domain of databases,
during 1990's [3].
Figure 1 depicts a generalized architecture of big data. Google formulated programming paradigms
like MapReduce and GFS, to cope up with the trials brought at the Internet by data administration and
interpretation. Besides, the load generated by sensors, clients, and additional worldwide reservoirs of data,
further powered the overwhelming streams of data that lacked certain amendment on the computing structure
and far-reaching data processing machine [4].

 ISSN: 2252-8776
IJ-ICT Vol. 8, No. 1, April 2019 : 39–49
40
Figure 1. An overview of big data
Practically, all the foremost establishments have initiated their individual developments concerning
the big data within a few former years, comprising Google, Facebook, Microsoft, EMC, Amazon, Oracle, and
IBM, etc. Likewise, several nation-wide governments have also paid abundant devotion to big data and made
millions of funds to initiate the Project regarding the Analysis and Advancement of the Big Data [5]. The
concluding objective of big data stands to deliver the productivity as some commercial resolutions that can
comfort a company to gain professional solutions.
For instance, any company can be benefited if they could understand that if client purchases “X”
then it is probable that he/she might also be interested in buying “Y”. This type of analysis at run-time can
greatly benefit by increasing business. The web accounts are analysed by the online sites offering human
interaction in order to propose some preferences to the users based upon their vested interests. Big data also
targets on remarkable reduction in expenses and necessary developments [6].
There are three main keys for big data, also known as 3 V’s of big data. [7]
i. Volume - Presently the data size is much larger in comparison to past data sizes, i.e. exceeding -
terabytes and peta bytes. The striking range and gradual surge in the data size tags it vigorously hard
to save and review by employing the conventional approaches. For instance, Facebook consumes
approximately 500-terabytes of data on a daily basis.
ii. Velocity - The utilization of the big data is must as it streams the data to obtain the optimum use of
its value for time restricted processes.
iii. Variety -Origination of the big data is primarily based on the diversity of sources. The Conventional
systems of databases were proposed to mark lower extents of classified data, smaller amount
updates or a steady and feasible data arrangement. However, the spatial data, 3-D data, audio-video,
and the cluttered manuscript, comprising account files and social media are also considered as big
data.
Big Data technology permits the collection and processing of large extents of data, including personal
information or information that can recognize an individual. Presently, the data has transformed as an
imperative constituent that could be analogous to real assets and individual resources. Generally, there are
five custom ways through which the big data can be used [8]. First, it can create information more crystal
clear and rapidly. Second, the establishments can assemble and examine further digital data, precisely. Third,
the utilization of such data can generate much more accurately personalized goods or facilities for consumers.
Fourth, pooled with the precise analytics and Data Discipline, the process of decision-making considerably
turns into more proficient. Fifth, it can be utilized to mend the succeeding generation of amenities and yields
for a company’sclient base. Currently, big data has been utilized in practically every single field [9]. Some of
the fields that are consuming big data services are defined below:
i. Retail: The foremost task of business industry is building client relationship with the associations or
organizations. The optimum way to grasp and dominate clients is conduct dealings and tactics
efficiently in order to procure back the unsuccessful goods and progression of the premium goods.
ii. Manufacturing: The companies can improve the superiority and efficiency of the manufactured
goods by minimizing the leftover with the awareness information delivered by big data. Several
companies are presently providing stress to analytics-based policy for resolving difficult and flexible
decision making.
iii. Education: Education completely examines the data occupied from the school faculty association
can create dominant impact on organizing endangered learners and observing the sufficient

IJ-ICT ISSN: 2252-8776 
Recent trends in big data using hadoop (Chetna Kaushal)
41
improvement of students. Assessment of the student’s development can be made with the school
faculty synchronization enhanced system.
iv. Healthcare: Miscellaneous patient records, treatment data and processes for therapy are
accomplished effectively with the awareness of information; health care providers can comprehend
and proficiently recover patient’s fitness.
v. Media/entertainment: From the past five years, the industry of social media/entertainment has
shifted to the digital means of production, recording, and circulation is currently accumulating
enormous amounts of users observing actions and the rich content.
vi. Life sciences: Nearly tonnes of information (measured in terra-bytes) are produced by lesser price
DNA sequencing which is required to be examined in order to scan the hereditary modifications and
possible proficiency of the cure.
vii. Video surveillance: Video surveillance is developing from CCTV toward IPTV recording systems
and capturing devices like cameras that are used by the organizations as per the need to analyse
patterns of activities and actions (enhancement of service and security).
viii. Transportation, utilities, services, telecommunication and logistics: At high rate sensor data is
generated from the GPS transceivers, smart meters and mobile devices (cell phones) which is then
used for optimizing the operations and find the relationship between the data which form relevant
information for business intelligence (BI) to make the appropriate decisions for different business
opportunities.
2. DATA MINING
Big data on cloud contains all the raw data which is gathered in clusters on the basis of their
relationship. But the user or organization never wanted to waste their time in gathering the data details and
creating structural information as it takes a lot of time. Hence, Data Mining is referred as taking out info from
vast groups of records of data. In other way, the process of data mining is to mine knowledge from the
database [10]. There is a vast quantity of data existing in IT Industry. Such data cannot be utilized further for
processing, unless that data is transformed into valuable info. It is indispensable to analyze enormous volume
of data and mine the valuable information from the data. Mining of the information is not only procedure
which is particularly required to be performed; there are also other processes that are involved in data mining
like Data Cleaning, Data Selection, Data Integration, Data Transformation, Data Mining, Data Presentation
and Pattern Evaluation is described in Figure 2 [11]. Once all these jobs are completely terminated, this
information could be adapted further in various applications as Fraud Exposure, Market Analysis, Science
Exploration and Control in Production etc. [12].
Figure 2. Data mining in Knowledge Discovery process [13]
Data mining, often referred to Knowledge discovery (KDD) involves mining of information or
knowledge as its primary and the utmost challenging and intriguing step [13]. Normally, data mining
discloses the intriguing patterns and inferences that are concealed covertly inside a large volume of
unanalysed or primary data, and the outcomes which are carried out may possibly support future observations
in the actual world.
Data mining has been exploited by an extensive variety of applications i.e. business, drug, science
and engineering. Although, the data mining is main phase in knowledge discovery process therefore it is also
used as a substitute for entire process of taking out useful info from databases.

 ISSN: 2252-8776
42
But still in technical environment (industry), in the database research and in media, data mining is
becoming extraprevalent. There are number of steps involved in the entire process of knowledge discovery
from databases which is described in Figure 2 [14]. The figure represents the sequence of the individual step
within the process and is briefly described in the text below:
− Data Integration − the multiple data from distinct sources are initially joined.
− Selection of data– the appropriate data as per the analysis of the task is selected from multiple data
sources.
− Data Pre-Processing − the noise and inconsistency of the data is eliminated.
− Data Transformation − data is converted or merged into such forms that are applicable for mining by
carrying out summary or blended procedures.
− Data Mining –intellectual procedures are adapted to abstract the patterns of data.
− Evaluation of Pattern− evaluations of data patterns which are abstracted.
− Knowledge Presentation –In Conclusion, knowledge is represented.
2.1. Challenges in Big Data Mining
The foremost challenges that arise in big data mining are briefly defined in the following points at
Table 1 [15]:
Table 1. Challenges in the Big Data Mining
Challenges Description
Shielding privacy and
confidentiality
Prime focus on generating the techniques that will never disclose the designs and also ensure safety and
privacy
Managing the
inadequate information
Absent values that relates to deficiency of features, is argued comprehensively for offline, static settings
Undefined data Most applications do not possess sufficient data for arithmetic procedures. Hence approaches are required to
handle undefined data values in a precise and quick way.
Diversity of data Social site is the most captivating imminent application of data stream clustering like video, images, text
and audio.
Synopsis and summaries Synopsis refers to compressed statistics arrangements which let data summarization for advance questioning
like the histograms, wavelets forms and samples define the enormous information in the compressed way.
Distributed streams In applications such as centralized results bring together interruptions in event recognition and response that
can create mining systems unsuccessful
Evaluation of data
stream procedures
Existing tools such as idea implication, restricted processing interval, authentication dormancy, multiple
stream structures are inadequate in the data stream databases due to certain problems
Independent and self-
diagnosis
Knowledge discovery in databases need the skills for prognostic self-diagnosis. A meaningful and beneficial
intellectual feature is diagnostics in situation of failure occurrence and also prognostic and consultative. The
evolution of these types of self-organizing, self-enhancing, and self-restoring systems is foremost challenge.
Merging offline and
online models
Real-time and batch learning are frequently considered as separated identities according to their action, but
their grouping might boost the data value. In lambda framework the two models can be combined for
planning big data models.
3. HADOOP
Hadoop is an open-source framework which permits to accumulate and run big data in a distributed
arrangement in the network of computers consuming modest programming models. This whole process
scales up from solitary servers to thousands of machines, collectively put forward local manipulation and
storing. Hadoop executes the applications via MapReduce algorithm, where on diverse CPU nodes; info is
sort out in parallel. In a nutshell, Hadoop framework is proficient to encourage applications that are qualified
of executing on the group of machines and all could deliver fully statistical interpretation for immense
volumes of data [16].
The application which is dependent on Hadoop framework runs in an environment which gives
distributed storage and computations on the group of machines in the network. Extension of Hadoop could be
numerous servers, each giving the native computation and storage service.
In Figure 3, Hadoop Architecture is defined which primarily includes subsequent modules [17]:
i. Hadoop Common: Hadoop common consist of libraries of Java and services needed by other
Hadoop elements. These libraries offer OS level abstractions, files system and comprises essential
Java libraries and scripts required to initialize Hadoop.
ii. Hadoop YARN: It is kind of structure for scheduling of job and cluster resource managing.
iii. Hadoop Distributed File System (HDFS): It is file architecture that offers right to use the application
data.
iv. Hadoop MapReduce: This is a system based on YARN for parallel processing of big sets of data.

IJ-ICT ISSN: 2252-8776 
43
Since 2012, the concept "Hadoop" repeatedly mentions to be the base units and to the variety of
other software sets that can be mountedbeside Hadoop, like Apache Hive, Apache Pig etc.
Figure 3. Architecture of Hadoop [16]
3.1. Hadoop Distributed File System (HDFS)
The Hadoop Distributed File System (HDFS) is centred on the Google File System (GFS) and offers
file system which is distributed in nature that is intended to execute on large group of computer in the
network in a consistent and error-receptive manner [18]. In contrast to the additional distributed systems,
HDFS is exceedingly fault indulgent and designed with low cost hardware. HDFS grasps very huge amount
of data and offers comfortable access. The files are stored across several machines in order to protect such a
huge data. These files are kept in a repetitive manner to recover data losses in the system in events of failure.
HDFS primarily adapts the master/slave design. In this design the master comprises a single
NameNode that maintains the metadata and slave comprises multiple DataNodes that preserve the original
data.
In the Figure 4, architecture of HDFS is shown and is divided into data nodes. A file in referred as
HDFS namespace is divided into a number of blocks. These individual blocks are kept in a class of
DataNodes. [19]. The DataNodes are responsible for the reading and writing procedure of the file system.
They further are responsible for the block formulation, termination, and duplication as per the instructions
provided by NameNode. The HDFS renders a shell similarly to many other file system (meta data)and a list
of instructions are prepared to communicate to the file system.
Figure 4. Architecture of HDFS [18]
4. CLUSTERING ALGORITHM
Clustering is the job of division of the population/data points in group number like data points in
similar group that are same as another data points in similar group as compared to the another groups. It can

 ISSN: 2252-8776
44
also be said that the objective of clustering is to separate the group in same traits and allocate them
in cluster form. Clustering can be classified into two algorithms, namely, K-mean and K-medoids clustering.
The explanation of the same is given.
4.1. K-means algorithm
The most general algorithm utilizes an iterative refinement method. Because of its ubiquity, it
frequently termed as K-means algorithm or as Lloyd’s algorithm. Consider K-mean initial set/ Centroids, the
algorithm has been divided into two steps [20]:
Assignment Step: Assigning every observation to the cluster having the closest mean [that is the
partition of the observation as per Voronoi diagram produced by the means.
Ri
(q)
= {aj: ‖aj − ni
q
‖≤ ‖aj − ni∗
q
‖for all i∗
= 1, … , l}
Update step: Compute the novel mean to be centroid of the cluster observation
ni
(q+1)
=
1
|Ri
q
|
∑ ajaj∈Ri
(q)
The K-means algorithm is believed to be meet when the assignment doesn’t change for long [21].
4.2. K-medoids algorithm
K-medoid algorithm is associated to the K-mean algorithm with the medoidshift algorithm. The K-
medoid and the K-mean algorithm known as Partitional algorithms. K-mean lessens the total squared error
and the K-medoids reduces the amount of dissimilarities among points labeled to be in cluster with the point
selected as the cluster centre. K-medoids selects the data points as the centres with respect to K-mean
algorithm. It is a partitioning method for clustering the data sets of m objects in k-clusters by K termed as
Priori. The effective tool to measure is Silhouette [22].
It may be more vigorous to noise and the outliers by means of k-means as it reduces a amount of
normal pair wise dissimilarities than squared Euclidean distance sum. The medoid by means of finite dataset
is the data point form the set having average dissimilarity to each data point is less means it is considered as
likely to the centrally located point set. General realization of k-medoid clustering is PAM (Partitioning
Around medoid) algorithm and is defined below [23]:
i. Initialize: Arbitrarily selected K of m data points as medoids.
ii. Assignment step: Connect every data point to the closest medoid.
iii. Update step: for every medoid and for every data point p, linked to n swap n and p and calculate the
total configuration cost. Choose the medoid p with less configuration cost.
5. CLASSIFICATION ALGORITHM
Classification has wide range of methods to categorize the data into the group of clusters.There is
utter need of the classification process as the huge volume of data is categorized into the group based on the
relation between the data objects. Hence, algorithms are required which has training data-sets inbuilt
according to human perception of data classification. Classification is a typical data mining methodthat is
dependent on machine learning [24]. Basically classification is needed to classify each object into a particular
class. Classification is further divided into Supervised and Unsupervised classification.
Supervised learning is in which the training set of precisely recognized dataset observations are
accessible. Whereas, in the unsupervised learning takes the chance itself by grouping data on the basis of
similar measures of inherent similarity.
There are numerous methods in the supervised learning however according to the previous studies
KNN is the best method for classification in the case of big data and give better results when used. Here after,
KNN algorithm is defined and how classification is done with the help of KNN algorithm is presented in
subsection [25].
5.1. KNN Algorithm
The K-nearest neighbour procedure (KNN) is a way for classification of entities on the basis of the
adjoining training specimens in feature space [26]. The prime intention of the k Nearest Neighbours (KNN)
process is- to use the database wherein the data are divided into a number of isolated classes to prognosticate
the classification of a new sample point. KNN classification distributes the data into test set and training sets.
Then the K nearest training set objects are originated for every single row of the test set, and the process or

IJ-ICT ISSN: 2252-8776 
45
task of classification is performed by predominance vote with connections which can be broken at any
moment.
Figure 5. Working steps of KNN algorithm
In the Figure 5, The K-nearest neighbour algorithm (KNN) is summarised as:
i. A +ve number k is stated, with a new sample
ii. The k items are selected from the database that are next to new sample
iii. The utmost mutual classification of selected entries is determined.
iv. Resulted Classification is offered to the new sample.
In KNN classification, the output is a class membership. An object is classified through thebulk vote
from the nearby neighbours, with entity being allocated to class most mutual among the entities k adjoining
neighbours. If k = 1, the object is assigned to class of that sole nearest neighbour. A peculiarity of KNN
algorithm is that its sensitivity to local structure of data [27].
Assume, training set D
i. Object to be tested x = (x_, y_),
ii. After that algorithm calculates the similarity between z and all training objects to conclude its
nearest-neighbour list i.e. Dz.
Training objects =(x, y) ∈ D
iii. x = data of a training object,
y= is its class.
iv. Similarly, x_ = data of the test object
y_= is its class
The classification of test object is done on the basis of majority class of its nearest neighbours which
is described in the equation below:
Majority Voting: y′
=
argmax
v
∑ I(v = yi), (xi, yi)ϵDz (1)
In the above equation;
v =class label
yi=class label for i th nearest neighbours
I (·)= indicator function which returns the value 1
if its argument = true andotherwise 0 is returned as a value.
An Example of the k-NN classification has been explained briefly along with Figure 6. The Figure 6
demonstrated that the test model (i.e. green coloured circle) can be classified either to first class of the blue
coloured squares or to the other class of red coloured triangles. If k = 3, (considering solid line circle) then
the test model is allocated to the second class as there are2 triangles inside the inner circle and only 1 square.
Whereas, if k = 5, (considering the dashed line circle), the test model is allotted to the first class since there
are 3 squares inside the outer circle and only 2 triangles. The allocation is based on the majority vote of its
neighbour [28].

 ISSN: 2252-8776
46
Figure 6. k-NN classification
Euclidean Distance can be calculated by using:
D(x, y) = √∑ (xi − yi)2n
i=1 (2)
K-Nearest Neighbour can be predicted by employing the following equation:
y =
1
k
∑ yi
x
i=1 (3)
In the above equation, yi = ith case of test model; y = outcome of the query point.
In classification problems, on a voting scheme the KNN predictions are based and the winner is
used to label the query.
The k-NN algorithm accuracy can be strictly degraded with theexistence of noisy features,
inconsistent feature scales etc. A lot of research effort is put into choosing or scaling features to improve the
performance of classification.
The accuracy level of the KNN algorithm’s end result can be calculated by using the following
equation.
Accuracy = (
No.of correctly classified examples
No.of examples
) × 100 (4)
Pseudo-code for k-nearest neighbour classification algorithm [29]
knumber of nearest neighbors
for each object X in the test set do
Calculate the distance D(X, Y) between X and every object Y in the training set
Neighborhoodthe k neighbours in the training set closest to X
X.classSelectClass(neighborhood)
end for
Sometimes full description of the performance of classification algorithm is required and detailed
conception is a table entitled as the name of confusion matrix. The rows denote the real class of the test cases,
whereas, columns symbolises the prediction of classifiers. The title confusion matrix arises from observation
where the algorithm gets confused. Assume the database contains 100 players from the women gymnasts,
basketball association and marathon. The evaluation of classifier is done with 10-fold cross validation. The
results of this test are as shown in Table 2 [29]:
Table 2. Results of 10-Fold Cross-Validation
Gymnasts Basketball Players Marathoners
Gymnasts 83 0 17
Basketball Players 0 92 8
Marathoners 9 16 75

IJ-ICT ISSN: 2252-8776 
47
The actual class of each example is denoted by rows; the class anticipated by our classifier is
denoted by columns. So taken example,
83 = correctly classified gymnasts
17 = misclassified as marathoners.
92= correctly classified basketball players
8 = misclassified as marathoners.
75= correctly classified marathoners
9 = misclassified as gymnasts
16 =misclassified as basketball players.
The confusion matrix diagonal represents instances which were classified correctly.
In this case the accuracy of the algorithm is:
83+92+75
300
=
250
300
= 83.33%
6. RELATED WORK
Wu, X. et al, presented a comprehensive study regarding the topmost 10 algorithms of data mining
[25]. The algorithms whose comprehensive approach was mentioned were: C4.5, SVM, k-Means, EM,
Apriori, AdaBoost, Naive Bayes, CART and kNN. These algorithms included all clustering, classification,
association analysis; statistical learning and last linking that were treated as the most significant topics in the
research of data mining. The impact of algorithms has been discussed; comparison was done on the basis of
which future forecast has been delivered. In the later year, Bakshi, K., et al, focused on analysis of
unstructured data which refers to the information which may does not contain previously defined data model
or was not suitable to fit in relational tables [29]. There were many methods to tackle the problem of
unstructured data. The methods shared mutual features of elasticity, high accessibility and scale-out. Map
Reduce in unification with Hadoop file system which is mainly distributed and H-Base database, part of
Apache Hadoop plan which helped in analysing the unstructured data. Priyadharsini, C., et al, presented an
extensive study on methods of data mining and also summary of database related to knowledge discovery
[11]. The main focus was on the issues related to the data mining. Rodríguez-Mazahua L. et al, presented a
review of Big Data works for identification of the chief problems, tools, application area and developing
styles of Big Data [15]. To meet the objective, authors have studied 457 papers to classify the theories related
to Big Data. This analyzed work offered related material to researchers regarding key working in study and
Big Data application in diverse practical areas. Later, Shikha Singh,D.et al, discussed the challenges that
expand the utility of large data though attempting to grasp the appropriate strategy to procure previous
knowledge from large data stack[2]. There was yet a dispute concerning the mechanisms and established
management structures which were inefficient with Big Data. It highlighted such documents and several new
technologies that reveal the challenges based on the idea of Big Data.
Alam,A., et al, defined the architecture and the challenges of HADOOP [17]. The main problem
area which has been mentioned was the iterative running of map-reduce processes from the beginning even in
little minor alteration in input. It was not a good approach as every time in the big data cloud the entries are
added or deletedin the bulk amount, the processing speed needs to be at its utmost level. In the solution,
caching scheme was described at small level which helped in managing the activities very well in map reduce
functions. Kesavaraj,G.,et al, specified the advantages and drawbacks of the different classification
algorithms and the best algorithms according to previous studies was KNN [24]. The average accuracy has
been calculated and the genetic algorithm has the best accuracy rate with 46.67%. The efficiency, precision,
accuracy, sensitivity of the classification algorithms has been compared and the neural has achieved the
second highest 62.8 after the back-propagation algorithm according to previous studies.Sokolova,M.,et al,
presented the analysis of the machine learning classification tasks which were binary, multi-class,
hierarchical and multi-labelled [28]. Different changes in the confusion matrix on various well-known
measures have been reviewed and compared. Gandhi et al. have implemented the existing K-mean, K-
medoids and the presented Modified K-medoid algorithms. The K-medoid is being executed has performed
better as compared to K-mean and existing K-medoids on huge data sets for execution time and clustering
quality in the experimental outcomes. The author has calculated Dunn’s index, total time, davies bouldin
index, Krzanowski and Lai, CalinskiHarabasz index for the verification of the modified K-medoids with
exisiting K-medoids and K-mean performance. It has been concluded from the result that the modified k-
medoids has performed better [31]. Arora et al. has obtained enhanced results of clustering by utilizing two
clustering algorithms by means of varied clusters being formed and by means of distance metric. Clustering
algorithms, like, K-mean and K-medoids ha been used on the dataset transaction 10k of KEEL. The input has
been arbitrarily dispended data points and accordingly, similarity clusters has been produced. It can be drawn

 ISSN: 2252-8776
48
from the research that the when the distance metric changes, the outcome of clustering algorithm
changes [31-34].
7. COMPARISON OF EXISTING WORK
This section depicts the comparison of the work of [30] and [31]. From the literature study, the
comparison has been made on the basis of execution time on big data clustering for different clustering
approaches. The results have been shown below in the form of Table 3 and Figure 7.
Table 3. Comparison of Execution Time of Big Data Clustering
K-mean (Execution time (secs)) K-medoid (Execution time (secs))
Gopi Gandhi and Rohit Srivastva [31] Preeti Arora et al. [30] Gopi Gandhi and Rohit Srivastva [31] Preeti Arora et al. [30]
0.2014 0.0358 0.2223 0.0384
Figure 7. Comparison of clustering approaches of [30] and [31]
Above table and graph depicts the comparison of big data clustering of execution time of [30] and
[31]. The comparison has been made on the basis of K-mean and K-medoid approaches. The author Gopi
Gandi and Rohit Srivastva has used Similarity index with K-medoids clustering technique to enhance the
performance of clustering. So, the execution time in their work is less as compared to the work of Preeti
Arora et al. As shown in the graph and table, the value of execution time for [31] is 0.2014 and for [30], it is
0.0358 for K-mean approach. Similarly, in case of k-medoids, the value in case of [31] is 0.2223 and for [30],
it is 0.0384. The blue bar in the graph is depicting the work of Gopi Gandhi and Rohit Srivastva and red bar
is depicting the work of Preeti arora et al. The X-axis is for the approaches being utilized for the comparison
and Y-axis is showing the values of the execution time in seconds.
8. CONCLUSION
An overview of big data is presented along with big data usages and several challenges that are
associated with big data. This paper covers the study on data mining and knowledge discovery in databases
(KDD) with all the steps that are involved in the KDD process. The issues related to the clustering techniques
in data mining are also discussed briefly. The complete architecture of Hadoop and HDFS is also studied and
discussed. For classification of the data, several traditional methods such as rule based, decision tree,random
forests, boosting, quadratic classifiers associated with classification are briefly studied and then KNN
classification algorithm is selected for the data mining and described in this paper.An example is taken to
prove the accuracy of KNN algorithm which is measured to be 83.33%. A comparison has been made on
clustering algorithms, namely, K-mean and K-medoid for execution time of the existing
work of [30] and [31].

IJ-ICT ISSN: 2252-8776 
49
REFERENCES
[1] Gupta R, Gupta S, Singhal A, “Big data: overview,” arXiv preprint arXiv:1404.4136. Apr, 2014
[2] Shikha Singh, D. and Singh, G, “Big Data: A Review,” International Research Journal of Engineering and
Technology (IRJET), pp. 822-824,2017.
[3] Bano S, Kulkarni RA, Damade MK, “ Big Data: An Emerging Technology towards Scalable System,”2015.
[4] Shah R, Katre NM, Srivastava K, “ Teradata: The global leader in Data Analytics,”.
[5] Aarnio T, “ Parallel data processing with MapReduce InTKK T-110.5190,” Seminar on Internetworking, 2009.
[6] Kim GH, Trimi S, Chung JH, “ Big-data applications in the government sector,” Communications of the ACM,
Vol.57, issue 3, pp.78-85,2014.
[7] Sagiroglu S, Sinanc D, “ Big data: A review,” InCollaboration Technologies and Systems (CTS), IEEE
International Conference, pp. 42-47, 2013.
[8] Villars RL, Olofson CW, Eastwood M, “ Big data: What it is and why you should care,” White Paper, IDC, 2011.
[9] Becker T, Curry E, Jentzsch A, Palmetshofer W, “ New Horizons for a Data-Driven Economy: Roadmaps and
Action Plans for Technology, Businesses, Policy, and Society.” InNew Horizons for a Data-Driven Economy, pp.
277-291, 2016.
[10] Chen MS, Han J, Yu PS, “Data mining: an overview from a database perspective,” IEEE Transactions on
Knowledge and data Engineering, Vol. 8, issue 6, pp. 866-83,1996.
[11] Priyadharsini.C, and Dr. Thanamani,A. S, “An Overview of Knowledge Discovery Database and Data Mining
Techniques,” International Journal of Innovative Research in Computer and Communication Engineering, pp.
1571-1578,2014.
[12] McCue C, “Data mining and predictive analysis: Intelligence gathering and crime analysis,” Butterworth-
Heinemann, 2014..
[13] Hafez AM, “Knowledge Discovery in Databases,”.
[14] Maimon,O., and Rokach,L, “Data mining and knowledge discovery handbook,” Springer, 2005.
[15] Rodríguez-Mazahua L, Rodríguez-Enríquez CA, Sánchez-Cervantes JL, Cervantes J, García-Alcaraz JL, Alor-
Hernández G, “A general perspective of Big Data: applications, tools, challenges and trends,” The Journal of
Supercomputing,Vol. 72, issue 8, pp.3073-113,2016.
[16] White T, “Hadoop: The definitive guide,” O'Reilly Media, Inc, 2012.
[17] Alam A, Ahmed J, “Hadoop architecture and its issues.” Computational Science and Computational Intelligence
(CSCI), Vol. 2, pp. 288-291,2014.
[18] Hanson JJ, “ An introduction to the Hadoop distributed file system,” IBM-United States, 2011.
[19] Shvachko K, Kuang H, Radia S, Chansler R, “ The hadoop distributed file system,” InMass storage systems and
technologies (MSST), pp. 1-10,2010.
[20] Teknomo, K, “K-means clustering tutorial,” Medicine, Vol. 100,issue 4, 2006.
[21] Rauf, A., Sheeba, S. M., Khusro, S., & Javed, H, “Enhanced k-mean clustering algorithm to reduce number of
iterations and time complexity,” Middle-East Journal of Scientific Research, Vol.12, issue 7, pp.959-963, 2012.
[22] Park, H. S., & Jun, C. H, “A simple and fast algorithm for K-medoids clustering,” Expert systems with
applications, Vol. 36, issue 2, pp. 3336-3341,2009.
[23] Zhang, Q., & Couloigner, I, “A new and efficient k-medoid algorithm for spatial clustering,” Computational
Science and Its Applications, pp. 207-224,2005.
[24] Kesavaraj G, Sukumaran S, “ A study on classification techniques in data mining,” Computing, Communications
and Networking Technologies (ICCCNT), pp. 1-7,2013
[25] Wu X, Kumar V, Quinlan JR, Ghosh J, Yang Q, Motoda H, McLachlan GJ, Ng A, Liu B, Philip SY, Zhou ZH,
“Top 10 algorithms in data mining,” Knowledge and information systems, Vol.14, issue 1, pp. 1-37,2008.
[26] Archana S, Elangovan K, “ Survey of classification techniques in data mining,” International Journal of Computer
Science and Mobile Applications, Vol. 2, issue 2, pp. 65-71,2014.
[27] Kumar R, Verma R, “ Classification algorithms for data mining: A survey,” International Journal of Innovations in
Engineering and Technology (IJIET), Vol. 1, issue 2, pp. 7-14,2012.
[28] Sokolova M, Lapalme G, “ A systematic analysis of performance measures for classification tasks,” Information
Processing & Management, Vol. 45, issue 4, pp. 427-37,2009.
[29] Bakshi K, “ Considerations for big data: Architecture and approach,” Aerospace Conference on IEEE, pp. 1-
7,2012.
[30] Arora, P., & Varshney, S, “Analysis of K-Means and K-Medoids algorithm for big data,” Procedia Computer
Science,Vol. 78, pp. 507-512,2016.
[31] Gandhi, G., & Srivastava, R, “Analysis and Implementation of Modified K-Medoids Algorithm to Increase
Scalability and Efficiency for Large dataset,” International Journal of Research in Engineering and Technology,
2014.
[32] Bagheri, H., & Shaltooki, A. A, “ Big Data: challenges, opportunities and Cloud based solutions,” International
Journal of Electrical and Computer Engineering, Vol. 5, issue 2, 2015.
[33] Padhy, R. P, “Big data processing with Hadoop-MapReduce in cloud systems,” International Journal of Cloud
Computing and Services Science, Vol. 2, issue 1, 2013.
[34] Almohsen, K. A., & Al-Jobori, H, “Recommender Systems in Light of Big Data,” International Journal of
Electrical and Computer Engineering, Vol. 5, issue 6, 2015.

06. 9534 14985-1-ed b edit dhyan

More Related Content

What's hot (20)

Similar to 06. 9534 14985-1-ed b edit dhyan (20)

More from IAESIJEECS (20)

Recently uploaded (20)

06. 9534 14985-1-ed b edit dhyan