8 VI June 2020

http://doi.org/10.22214/ijraset.2020.6243
 International Journal for Research in Applied Science & Engineering Technology (IJRASET)
ISSN: 2321-9653; IC Value: 45.98; SJ Impact Factor: 7.429
Volume 8 Issue VI June 2020- Available at www.ijraset.com

Big Data in the Machine Learning Techniques

Perspective - A Review
Olawale Adepoju1, Devaraj Verma C.2
1, 2
Department of Computer Science and Engineering, Jain (Deemed to be University), India.

Abstract: In this information age, the decision-makers have been able to access enormous quantities of data, and every day a
massive stock of petabytes and zeta bytes of data is generated using today’s digital services and modern technology such as the
Web, social networking sites, internet of things and cloud computing.
Machine learning techniques for the interpretation and analysis of these data are applied quite effectively. There are however
various machine-learning strategies for analysing data, all of which do not work on every data on the same basis. In this paper,
we have presented a survey about some machine learning techniques and discussed their application along with the steps
involved for it.
Keywords: Machine Learning, Data Mining, Big Data, Internet of things, Cloud Computing

I. INTRODUCTION
Imagine a world with no preservation of information where every individual perspective, every interaction, or angle retrievable
about a person or organization is actually lost following use. This will lead to a lack of capacity for companies to collect valuable
data and analyse them, as well as new prospects and priorities. This has proved to be fundamental to routine workouts, everything
from details on consumers related to sales, to the workplace, and so forth. Data or information is a structure under which every
company thrives. Currently, consider the level of details and the growth of information and data collected these days through
advances and web technologies. When storage capacities and data collection techniques have increased, monumental knowledge
initiatives have proved to be efficiently available. The solution needs to be analysed and given because of the rapid increase of
these data, in order to process and derive value and information from such data. Policy-makers will need useful insights into other
complex and fast-changing data from daily transactions to consumer experiences and social network data.
This value can be given by big data analysis, which uses state-of-the-art analytical techniques for big data. The purpose of this
review is to breakdown a portion of the numerous algorithms for machine learning that can be related to big data, and the
possibilities provided by the use of large-data analysis in different option areas. The machine learning system encompasses the pre-
processing, learning, and evaluation methods.

II. LITERATURE REVIEW

A. Big Data
A vast variety of researchers around the world use the term big data, which reveals remarkably large (structured or unstructured)
datasets. Such data are too big and moving rapidly and modern databases are not fit to manage such massive data. A large number of
data sets are obtained from different fields such as sensors, transactional applications, networks, and social media, etc.
Big data has stimulated the future of information technology [1], focused mainly on the third platform, primarily on big data, cloud
computing, internet, and social enterprises, from the software and information technology background. Such data are available in
petabytes and beyond in structured, semi-structured, and unstructured formats. It is classified between 3V and 4V. 3V is about
distance, pace, and variety. By understanding the numerous V's associated with the number, speed, variation, and veracity, the big
data phenomenon can be clearly represented.
Three V’s were described by Doug Laney for big data [2]. The length, pace, and variety come from this.
1) Volume: This explains the quantity of data per second accumulated from Exabyte to zeta bytes in various tools, such as
transaction processes, community media, social media, etc. Since 2005 and 2017 more than 130 data from Exabyte have been
generated and Internet activities have increased.
2) Speed: This function reflects the pace at which data are collected and analyzed to meet demands. The strong and continuous
inflows of knowledge, data are collected in real-time and accurately, making it useful and precise for researchers to take
decisions.

© IJRASET: All Right s are Reserved 1488

 International Journal for Research in Applied Science & Engineering Technology (IJRASET)
ISSN: 2321-9653; IC Value: 45.98; SJ Impact Factor: 7.429
Volume 8 Issue VI June 2020- Available at www.ijraset.com

3) Variety: This word applies to the different data we may use. Wide range: The structure, structure, or semi-structure may be
used. Data are typically stored as tablets or databases, although there have been problems with data in recent times. Some
information is given by e-mails, records, photographs, video, tracking sensors, and much more representing unstructured data.
4) Veracity: This explains the accuracy of knowledge. In other words, the data refers to distortion, noise, anomalies, etc.

Fig. 1 The 4V’s which characterize Big Data

Massive volume, speed, flexibility, and accuracy of information through various traditional and computational insightful policies are
the primary target of large-scale analyses of information [3]. Gandomi and Haider investigated part of these extraction techniques
for data acquisition [4].

B. Machine Learning
Machine learning is a collaborative area of study that brings together insights into several parts of science, including machine
intelligence, statistical analysis, and IT. Machine learning is alluded to as the gaining from previous interaction with a few
undertakings and implementation steps. The motivation behind the algorithms is to learn "without specifically programming," as
Arthur Samuel described in 1959 [5], from the current knowledge. The machine learning approach enables users to uncover a
hidden framework and create independent predictions of large data sets.
Such predictions are derived from previous estimates and from repeatable results. A few instances of the day by day life machine
learning applications incorporate:
1) Self-Driving Vehicles: Such vehicles contain sensors that identify things from a vast region in every direction. The Car has tools
to evaluate and process all data collected on the road for secure navigation [6, 7]. Such an instance is indeed the perfect
illustration of machine learning implementations.
2) In Recommender systems also, machine learning has made waves. For example Amazon or Netflix Recommender Mechanisms.
The recommendation system in Netflix depended on restricted Boltzmann machine and grid factoring [8].
3) Also in Fraud detection, the machine has been very useful, it is one of the most important aspects nowadays. All e-business face
this challenge of fraud and network intrusion.

Machine learning problems are basically classified under three classes. They are:
a) Supervised learning involves labeled information to be trained. Supervised learning involves labeled information to be trained.
For every information labeled includes data and ideal value for the target yield. The learning algorithm breaks down the trained
information and provides a deducted capacity that can be used to map new qualities.
b) For example, clusters evaluation is unlabeled information in the unsupervised learning process. The result or consequence of
the data is unknown. It is used for clustering (grouping) problems, for anomaly detection (in banks for irregular transactions),
where links between the data are required.
c) The third class, reinforcement, enables an agent to take its result from the input to the ties to the external situation [9].
Reinforced learning is a kind of learning in which individual trains himself to benefit from its experience so that its reward can
be amplified. The agent has no guidance of options, however on the off chance that the order is successfully implemented, it is
compensated at that stage and else, it receives a negative reward. When the agent has a comparative situation, it must determine
if the option depends on the experience of the past. The compensation is numerical validation and tags for the success of an
operation.

© IJRASET: All Right s are Reserved 1489

 International Journal for Research in Applied Science & Engineering Technology (IJRASET)
ISSN: 2321-9653; IC Value: 45.98; SJ Impact Factor: 7.429
Volume 8 Issue VI June 2020- Available at www.ijraset.com

Fig. 2 Reinforcement Learning Curve

Through an information processing perspective, both the supervised and unsupervised training system is favoured for information
analysis, and reinforcement strategies are favoured for improving specific choice-making [10].
The outcomes might change the estimation for the selected measure of training. Structuring a learning framework in machine
learning includes four plan decisions:
 Preparing Data
 Choose target feature
 Assign representation
 Usage of the correct algorithm of learning

III. METHODOLOGY
While information continues to expand rapidly every day, various insightful learning approaches are being introduced in order to
respond to some key data prediction analytical problems. This section demonstrates clearly a variety of machine learning
approaches for the big data study.
Several learning techniques are there in machine learning for analysis of big data, some of these techniques’ are but not limited to
Logistic Regression, Support Vector Machine, Linear Regression, K-Means Clustering, Naïve Bayes, and Artificial Neural
Networks.

A. Logistic Regression
The logistics regression is the sufficient statistical analysis for the dichotomous (binary) dependent variables. It is predictive
analysis, as in all regression analyses, and also used to characterize data and illustrate the correlation between a binary variable
dependent and one or more nominally independent variables, interval variables, or ratio rates. The relationship between both
variables is not seen as a straight line by logistic regression. Logistic regression uses the natural logarithm function instead to
determine the relationship between the variables and uses testing data to determine the coefficients.
The logistic regression is based on binary data, where the occurrence happens or the event does not happen. The function forecasts
future outcomes. In light of other function x, it attempts to figure out whether or not an occurrence occurs. Then y can be either 0 or
1. If the occurrence occurs, y is assigned the value 1, then y can be either 0 or 1. If the occurrence occurs, y is assigned the value 1.
In the logistic regression, a model corresponding to data points is found using the sigmoid function. The method gives the data
model an 'S 'curve. The curve is between 0 and 1, and when y is binary, it is simple to apply.
There is a threshold of 0.5 for the curve, it is assumed that if the value is higher than 0.5 it is classified as 1 and else classified as 0.
Following is the sigmoid function:
yi = ( )

The Sigmoid function is used to find a binary classification probability. In this equation, y is the probability of the output, and z is th
e log - odds of the example.
z = b + m1 x1 + m2x2 + m3x3 + ............ mnxn
Thus b is the intercept of linear regression, m is the values and bias weighted, and x the values shown.
The sigmoid function predicts the likelihood of a certain outcome.

© IJRASET: All Right s are Reserved 1490

 International Journal for Research in Applied Science & Engineering Technology (IJRASET)
ISSN: 2321-9653; IC Value: 45.98; SJ Impact Factor: 7.429
Volume 8 Issue VI June 2020- Available at www.ijraset.com

Fig. 3 Diagram of Logistic Regression Curve

B. Support Vector Machine

The essential idea driving Support Vector Machine (SVM) is a hyper-plane used in the decision. By plotting any acquired data value
in an n-dimensional space or graph, a support vector algorithm is performed. The total number of data features present here is "n."
The value of each data is displayed as a different graph coordinate. Svm is an example of supervised learning techniques that can be
implemented in problems of classification and regression. Practically, the grouping procedure isn't straightforward, and, when it
depends on the partition utilizing diverse lines it is known as hyper plane classifiers.
The optimum hyper plane is chosen according to the line-distance. Svm Vector help divides the class from one side to the other. The
difference is called the margin and the margin is referred to as vectors of support.

Fig. 4 Support Vector Machine Hyper-plane

C. Linear Regression
This algorithm defines a line that is best matched by determining different metrics for all the inputs variables known as a
coefficient(s). This line is better adjusted between inputs (X), output (Y). The main purpose is to get a line that fits the data the best.
That most suitable line is the one with a low total forecast error (every data point). Error is the range from the regression line points.
Y (predicted) = b0 + b1 * x
Y (predicted) must be forecasted in view of the data x and discover the values of the b0 and b1 coefficients as part of the linear
regression. Another is a predictor and the other is a variable dependent. Yet it does not follow a probabilistic relationship, which is
measurable. In the absence of a variable that can be precisely expressed, the relation between the two variables is called
probabilistic. For example, a direct, variable-based mathematical solution for conventional least squares and optimization of
gradient descent can be used to take linear regression from data. For example, a direct, variable-based mathematical solution for
conventional least squares and optimization of gradient descent can be used to take linear regression from data.

© IJRASET: All Right s are Reserved 1491

 International Journal for Research in Applied Science & Engineering Technology (IJRASET)
ISSN: 2321-9653; IC Value: 45.98; SJ Impact Factor: 7.429
Volume 8 Issue VI June 2020- Available at www.ijraset.com

Fig. 5 A Regressor Line on Data

D. Artificial Neural Network

A computational model is a network of artificial neurons (ANN), based on biological neural network architectures and features.
Data that passes via the network impacts the ANN structure as the neural network, based on the input and output, improves learning.
Artificial Networks (ANN) are neuronal systems with a multi-layer relation.
These are made up of an input layer, several hidden layers, and an output layer. Each node in a layer is connected to each other node
in the subsequent layer. By increasing the number of hidden layers, we are increasing the network.

Fig. 6 A representation of Artificial Neural Network

IV. CONCLUSIONS
Machine Learning is vital in tackling the challenges facing a generation of data, data retrieval, and large data processing, which is
mainly aimed at turning knowledge into a real impetus for improving businesses and consistent analysis. We talked about big data in
this article and also displayed a few machine learning attributes. It has shown several applications of machine learning in our
everyday life.
The potential reach of data analysis is to see how Machine Learning is made easier for non-specialists in multiple environments to
communicate with an alternative type of knowledge.

V. ACKNOWLEDGMENT
The authors are grateful to Jain University, Department of Computer Science and Engineering, for their in-valuable support and
feedback on this work.

© IJRASET: All Right s are Reserved 1492

 International Journal for Research in Applied Science & Engineering Technology (IJRASET)
ISSN: 2321-9653; IC Value: 45.98; SJ Impact Factor: 7.429
Volume 8 Issue VI June 2020- Available at www.ijraset.com

REFERENCES
[1] X. Jin, B. W.Wah, X. Cheng and Y. Wang, Significance and challenges of big data research, Big Data Research, 2 (2) (2015), pp.59-
64. https://doi.org/10.1016/J.BDR.2015.01.006.
[2] Doug Laney, 3D Data Management: Controlling Data Volume, Velocity, and Variety, META Group Research Note, 2001, 6: 70.
[3] M. K.Kakhani, S. Kakhani and S. R.Biradar, Research issues in big data analytics, International Journal of Application or Innovation in Engineering &
Management, 2 (8) (2015), pp. 228-232.
[4] A. Gandomi and M. Haider. Beyond the hype: Big data concepts, methods, and analytics, International Journal of Information Management, 35 (2) (2015), pp.
137-144. https://doi.org/10.1016/j.ijinfomgt.2014.10.007
[5] Big Data History and Current Considerations, http://www.sas.com/en_us/insights/big-data/what-is-big-data.html
[6] Machine Learning, https://en.wikipedia.org/wiki/Machine_learning
[7] Poczter, S. L., & Jankovic, L. M. (2013). The Google Car: Driving Toward a Better Future? Journal of Business Case Studies (JBCS), 10(1), 7-14.
https://doi.org/10.19030/jbcs.v10i1.8324
[8] Xavier Amatrian. How does the Netflix movie recommendation algorithm work? 2014, https://www.quora.com/How-does-the-Netflix-movie-
recommendation-algorithm-work (accessed 27 December 2014)
[9] C.M. Bishop, Pattern recognition and machine learning, Springer, New York, 2006.
[10] J. Qui, Q. Wu, G. Ding, Y. Xu and S. Feng, A survey of machine learning for big data processing. EURASIP Journal on Advances in Signal Processing,
Springer, 2016, 67 (2016). https://doi.org/10.1186/s13634-016-0355-x.

© IJRASET: All Right s are Reserved 1493