Chapter 2

Big Data Analytics

and Algorithms
Alok Kumar, Lakshita Bhargava, and Zameer Fatima

Contents
2.1 Introduction...............................................................................................20
2.2 Big Data Analytics .....................................................................................20
2.3 Categories of Big Data Analytics................................................................21
2.3.1 Predictive Analytics ........................................................................23
2.3.2 Prescriptive Analytics......................................................................25
[Link] How Prescriptive Analytics Works....................................25
[Link] Examples of Prescriptive Analytics....................................25
[Link] Benefts of Prescriptive Analytics ......................................25
2.3.3 Descriptive Analytics ......................................................................26
2.3.4 Diagnostic Analytics.......................................................................26
[Link] Benefts of diagnostic analytics .........................................26
2.4 Big Data Analytics Algorithms...................................................................26
2.4.1 Linear Regression............................................................................28
[Link] Preparing a Linear-Regression Model ...............................29
[Link] Applications of Linear Regression.....................................30
2.4.2 Logistic Regression .........................................................................30
[Link] Types of Logistic Regression .............................................31
[Link] Applications of Logistic Regression...................................32
2.4.3 Naive Bayes Classifers....................................................................33
[Link] Equation of the Naive Bayes Classifers ............................33
[Link] Application of Naive Bayes Classifers.............................. 34

DOI: 10.1201/9781003175711-2 19
20  Big Data Analytics

2.4.4 Classifcation and Regression Trees................................................ 34

[Link] Representation of CART Model ...................................... 34
[Link] Application of Classifcation and Regression Trees ...........35
2.4.5 K-Means Clustering ........................................................................35
[Link] How K-Means Clustering Works ......................................36
[Link] Te K-Means Clustering Algorithm..................................36
[Link] Application of K-Means Clustering Algorithms ................36
2.5 Conclusion and Future Scope.....................................................................37
References ...........................................................................................................37

2.1 Introduction
Tere is no denying the fact that the digital era is on the horizon, and it is here to
stay. In this digital era, a shift is occurring from an industry-based to an information-
based economy, which has caused a large amount of data to be accumulated with a
mindboggling increase every single day. It is estimated that by 2025 we will be gener-
ating 463 exabytes of data every day. Tis staggering amount of data available is both
a boon and a curse for humanity. Improper handling of data can lead to breaches of
privacy, an increase in fraud, data loss, and much more. If handled properly, a tre-
mendous growth and enhancement in technology can be achieved. Te traditional
methods of handling and analyzing data like storing data in traditional relational
databases usually perform very poorly in handling big data, the reason being the sheer
size of the data. Tis is where the power of big-data analytics comes into full swing.
Te key highlight and main contributions of the chapter include

 Te main idea behind writing this chapter is to provide a detailed and struc-
tured overview of big-data analytics along with various tools and technology
used in the process.
 Te chapter provides a clear picture of what big-data analytics is and why it is
an extremely important and dominant technology in the current digital era.
 We have also discussed diferent techniques of big-data analytics along with
their relevance in diferent scenarios.
 A later section of the chapter focuses on some of the most popular and cut-
ting-edge algorithms being used in the process of big-data analytics.
 Te chapter concludes with a fnal section discussing the shortcomings of
current data analytics techniques, along with a brief discussion of upcoming
technologies that can bridge the gaps present in current techniques.

2.2 Big Data Analytics

Big‑data analytics in very simple terms is the process of finding meaning-
ful patterns in a large seemingly unorganized amount of data. The primary
 Big Data Analytics and Algorithms  21

goal of big-data analysis is always to provide insights into the source that is
responsible for the generation of data. These insights can be extremely valu-
able for companies to understand the behavior of their customers and how well
their product is working in the market. Big-data analytics is also extensively
used for revealing product groupings as well as products that are more likely
to be purchased together. A mindboggling real-world example of this is the
‘diaper-beer’ product association found by Walmart upon analyzing its con-
sumer’s data. The finding suggested that working men tend to purchase beers
for themselves and diapers for their kids together when coming back home
from work on Friday night. This led Walmart to put these items together,
which saw an increase in the sales of both the items. This finding gives a clear
demonstration of the power of big-data analytics for finding product associa-
tions, as by using classical product-association techniques it is nearly impos-
sible to find such a bizarre correlation. To get a better understanding of how
the process of big-data analytics works in the real world, let’s take an example
of how an ecommerce company can leverage the power of big-data analyt-
ics to increase the sales of their product. In this example, we would consider
the broad analysis of two categories of data, data generated by the users in
the course of purchasing a product and data generated in after-sales customer
service. Big-data analytics techniques like market-basket analysis, customer-
product analysis, etc. can be used in the first kind of dataset to find asso-
ciations like product–product association, customer–product association, or
customer–customer association. These findings can be used by the company
to improve its product-recommendation system as well as product placement
on its portal. Similarly, the results obtained after analysis of after-sales data
like customer care phone calls, complaint emails, etc. can be used for train-
ing customer-care personnel or even in the development and improvement
of smart chatbots. These factors combined can increase the overall customer
satisfaction, which can boost the sales number and also help in new-customer
acquisition. A surface-level picture of the process is provided in Figure 2.1.
Big-data analytics also have found widespread application in the field of medi-
cal science. Various data-mining and analytics techniques have been used in a
variety of medical applications like disease prediction, genetic programming,
patient data management, etc. [1–3]. Data analytics can also be used in edu-
cational sectors to analyze students; data and generate better frameworks for
enhancing their education [4–5].

2.3 Categories of Big Data Analytics

Big-data analytics is usually classifed into four main categories as shown in
Figure 2.2. In this section, we will be looking into each of these categories in detail
as a separate subsection.
 22

Big Data Analytics
Figure 2.1 Levering Big-Data Analytics in An Ecommerce Company.
 Big Data Analytics and Algorithms  23

Figure 2.2 Categories of Big-Data Analytics.

Figure 2.3 Process of Predictive Analytics.

2.3.1 Predictive Analytics

Predictive analytics is a variation of big +-data analytics that is used to make predic-
tions based on the analysis of current data. In predictive analytics, usually historical
and transactional data are used to identify risks and opportunities for the future.
Predictive analytics empowers organizations in providing a concrete base on which
they can plan their future actions. Tis allows them to make decisions that are
more accurate and fruitful compared to the ones taken based on pure assumptions
or manual analysis of data. Tis helps them in becoming proactive and forward-
looking organizations. Predictive analytics can even be extended further to include
a set of probable decisions that can be made based on the analytics obtained during
the process. Te whole process of predictive analytics can be broken down into a set
of steps as shown in Figure 2.3.
24  Big Data Analytics

Steps involved in predictive analytics process:

1. Defne the project—Te frst and one of the most important steps in the
process of predictive analytics is defning the project. Tis step consists of
identifying diferent variables like scope and the outcome as well as identify-
ing the dataset on which predictive analytics needs to be executed. Tis step
is extremely crucial as it lays down the foundation for the whole process of
data analytics.
2. Data collection—Data is the most fundamental piece of every data-analytics
process; it’s the same when it comes to predictive analytics. In the data-collec-
tion stage organizations collect various types of data through which analytics
can take place. Te decision to determine the type of data that need to be
collected usually depends on the desired outcome of the process established
during the project defnition stage.
3. Data analysis—Te data analysis stage comprises cleaning, transforming,
and inspecting data. It is in this stage that patterns, correlations, and useful
information about the data are found.
4. Statistics—Tis is a kind of intermediate stage in which the hypotheses and
assumptions behind the model architecture are validated using some exist-
ing statistical methods. Tis step is very crucial as it helps in pointing out
any faws in the logic and highlights inaccuracies that may plague the actual
model if unnoticed.
5. Modeling—Tis stage involves developing the model with the ability to automat-
ically make predictions based on information derived during the data-analytics
stage. To improve the accuracy of the model, usually a self-learning module is
integrated, which helps in increasing the accuracy of the model over time.
6. Deployment—In the deployment stage, the model is fnally deployed on a
production-grade server, where it can automatically make decisions and send
automated decision reports based on that. It can also be exposed in the form
of an application programming interface (API), which can be leveraged by
other modules while abstracting the actual complicated logic.
7. Monitoring—Once the deployment is done it is advisable to monitor the
model and verify the predictions done by the model on actual results. Tis
could help in enhancing the model and rectifying any minor or major issues
that could cripple the performance of the model.

Predictive analytics is being used extensively to tackle a wide variety of problems

ranging from simple problems like predicting consumers’ behavior on the ecommerce
platforms to highly sophisticated ones like predicting the chance of occurrence of a
disease in a person based on their medical records. With the advancement in the feld
of data analytics, the accuracy of predictive analytics models has increased exponen-
tially over the decade, which has enabled their uses in the feld of medical science.
Maryam et al. have discussed various predictive analytics techniques for predicting
 Big Data Analytics and Algorithms  25

Drug Target Interactions(DTIs) based on analysis of standard datasets [6]. Shakil et

al. have proposed a method for predicting dengue disease outbreaks using a predictive
analytics tool Weka [1].

2.3.2 Prescriptive Analytics

Prescriptive analytics is a branch of data analytics that helps in determining the
best possible course of action that can be taken based on a particular scenario.
Prescriptive analytics unlike predictive analytics doesn’t predict a direct outcome
but rather provides a strategy to fnd the most optimal solution for a given scenario.
Out of all the forms of business analytics, predictive analytics is the most sophis-
ticated type of business analytics and is capable of bringing the highest amount of
intelligence and value to businesses [7].

[Link] How Prescriptive Analytics Works

Prescriptive analytics usually relies on advanced techniques of artifcial intelli-
gence, like machine learning and deep learning, to learn and advance from the
data it acquires, working as an autonomous system without the requirement of any
human intervention. Prescriptive-analytics models also have the capability to adjust
their results automatically as new data sets become available.

[Link] Examples of Prescriptive Analytics

Te power of prescriptive analytics can be leveraged by any data-intensive business
and government agency. A space agency can use prescriptive analytics to determine
whether constructing a new launch site can endanger a species of lizards living
nearby. Tis analysis can help in making the decision to relocate of the particular
species to some other location or to change the location of the launch site itself.

[Link] Benefts of Prescriptive Analytics

Prescriptive analytics is one of the most efcient and powerful tools available in the
arsenal of an organization’s business intelligence. Prescriptive analytics provides an
organization the ability to:

1. Discover the path to success—Prescriptive-analytics models can combine

data and operations to provide a road map of what to do and how to do it
most efciently with minimum error.
2. Minimize the time required for planning—Te outcome generated by pre-
scriptive-analytics models helps in reducing the time and efort required by
the data team of the organization to plan a solution, which enables them to
quickly design and deploy an efcient solution
26  Big Data Analytics

3. Minimize human interventions and errors—Prescriptive-analytics models

are usually fully automated and require very few human interventions, which
makes them highly reliable and less prone to error compared to the manual
analysis done by data scientists.

2.3.3 Descriptive Analytics

Descriptive analytics answers the question of what has happened. Te process of
descriptive analytics uses a large amount of data to fnd what has happened in a business
for a given period and also how it difers from another comparable period. Descriptive
analytics is one of the most basic forms of analytics used by any organization for get-
ting an overview of what has happened in the business. Using descriptive analytics on
historic data, decision-makers within the organization can get a complete view of the
trend on which they can base their business strategy. It also helps in identifying the
strengths and weaknesses lying within an organization. Being an elementary form of
analytics technique, it is usually used in conjunction with other advanced techniques
like predictive and prescriptive analysis to generate meaningful results.

2.3.4 Diagnostic Analytics

Te branch of diagnostic analytics comprises a set of tools and techniques that
are used for fnding the answer to the question of why certain things happened.
Diagnostic analytics takes a deep dive into the data and tries to fnd valuable hid-
den insights. Diagnostic analytics is usually the frst step in the process of business
analytics in an organization. Diagnostic analytics, unlike predictive or prescrip-
tive analytics, doesn’t generate any new outcome; rather, it provides the reasoning
behind already known results. Techniques like data discovery, data mining, drill-
down, etc. are used in the process of diagnostic analytics.

[Link] Benefts of diagnostic analytics

Diagnostic analytics allows analysists to translate complex data into meaningful
visualizations and insights that can be taken advantage of by everyone. Diagnostic
analytics also provides insight behind the occurrence of a certain result. Tis insight
can be used to generate predictive- or prescriptive-analytics models.
A comparison of all these four analytics processes along with the critical ques-
tion answered by each one of them is shown in Table 2.1 and Figure 2.4 respectively.

2.4 Big Data Analytics Algorithms

In the current digital era, data is the new gold. Every organization nowadays under-
stands the importance of having a stockpile of data at its disposal. Companies like
Google, Microsoft, and Facebook are dominating the modern era, and a big credit
 Big Data Analytics and Algorithms  27

Table 2.1 Comparison of Different Categories of Data Analytics

Category of
classifcation Predictive Prescriptive Descriptive Diagnostic

Source of Uses historical Uses Uses historical Uses

data data historical data historical
data data

Data Fills in gaps in Estimates Reconfgures Identifes

manipulation available data outcomes data into anomalies
based on easy-to-read
variables format

Role of Creates data Offers Describes Highlights

analytics models suggestions the state of data trends
about business
outcomes operation

Technique Forecasts Uses Learns from Investigates

used potential algorithms, the past underlying
future machine issues
outcomes learning,
and AI

Critical Answers ‘What Answers ‘If, Answer ‘What Answer

question might then questions’ ‘Why
answered happen?’ questions’ questions’

Figure 2.4 Critical Questions Answered by Different Analytics Techniques.

28  Big Data Analytics

Figure 2.5 Big-Data Analytics Algorithms.

for that goes to the mammoth data stores they have at their disposal. Having such
huge data stores at their disposal has enabled these companies to push the bound-
aries of technological advancement in a way that was never seen before. A burning
example that exhibits the power of data and what can be achieved through its
proper analytics is Google Maps. Built on top of data pipelines containing a huge
amount of dynamic and diverse data collected by Google from multiple sources, it
is a piece of technology that seems like something straight from the future.
But having data alone is not sufcient. Data on its own is useless and becomes
meaningful only when proper analysis of that data is done. With an unprecedented
increase in the amount of data generated in the last couple of years, it has become
more necessary now than ever to have fast and efcient data-analytics algorithms
at our disposal as the classical methods of data analysis using graphs or charts are
simply not enough to keep up with this huge amount of data otherwise also known
as Big Data. To solve this problem, data scientists all over the world have devel-
oped and are in the process of developing new advanced algorithms for analyzing
big data efciently. To discuss all of these algorithms is beyond the scope of this
chapter, hence we will keep our focus on the fve most popular big-data analytics
algorithms that usually form the basis of the majority of high-performance analyt-
ics models. Tese algorithms are shown in Figure 2.5 and discussed afterward.

2.4.1 Linear Regression

Linear regression is a kind of statistical test performed on a dataset to defne and fnd
the relation between considered variables [8]. Linear regression is one of the most
popular and frequently used statistical analysis algorithms. Being a very simple yet
extremely powerful algorithm for data analysis, it is used by data scientists exten-
sively for designing simple as well as complicated analytical models.
Linear regression, as the name suggests, is a simple linear equation that combines
the input values (x) and then generates the solution as a predicted output (y). In
the linear-regression model, a scale factor is assigned to each of the input values or
 Big Data Analytics and Algorithms  29

independent variables, which is also known as a coefcient and is symbolized using

the Greek letter Beta (˜). An extra coefcient, also known as intercept or bias coef-
fcient, is added to the equation, which provides an additional degree of freedom
to the line. If the linear-regression equation contains a single dependent variable
(y) and a single independent variable (x), it is known as univariate regression and is
represented by equation 2–1:

y = ˜1 * x + ˜0 (2–1)

y = dependent variable
x = independent variable
β1 = scale factor
β0 = bias coefcient

Te regression model with more than one independent variable is known as multi‑
variate regression. In a multivariate-regression model, an attempt is made to account
for the variation of independent variables in the dependent variable synchronically
[9]. Te equation of multivariate regression is an extension of univariate regression
and is represented in equation 2–2:

y = ˜0 + ˜1 * x1 + ˜ + ˜n * xn + ° (2–2)

y = dependent variable
x = independent variable
(˜1 − ˜n ) = scale factor
˜0 = bias coefcient
° = error

[Link] Preparing a Linear-Regression Model

Preparing a linear-regression model, also known as model training, is the pro-
cess of estimating the coefcients of the equation to fnd the best-ftting line for
our dataset. Tere are several methods for training a linear-regression model. In
this section, we will be discussing three of the most commonly used methods
among them.

1. Simple Linear Regression—Simple linear regression is a technique for

training linear-regression models when there is only one input—or, better
to say, only one independent variable—in the equation. In the method
of simple linear regression, model statistical properties from the data like
mean, standard deviation, correlations, and covariance are calculated,
which are used for estimating the coefcients and hence fnding the best-
ftting line.
30  Big Data Analytics

2. Least Square—Te method of least square is used when there are multiple
dependent variables and an estimation of the values of the coefcients is
required. Tis procedure seeks to attenuate the sum of the squared residu-
als. Te method suggests that, for a given regression curve, we can calculate
the space from each datum to the regression curve, square it, and determine
the sum of all of the squared errors together. Tis is often the value that the
method of least squares needs to attenuate.
3. Gradient descent—Te method of gradient descent is used in the scenario
when there are one or more inputs and there is a requirement for optimiz-
ing the value of the coefcient, which is done by an iterative minimization
of the error of the model on training data. Te algorithm starts by assigning
random values to every coefcient. Calculating the sum of squared errors for
all pairs of input and output values is the next step in the process of gradient
descent. A learning rate is associated, which acts as a multiplier with which
the value of coefcients are updated with the goal of minimizing the error.
Tis process gets terminated when either minimum-squared sum has been
achieved or any further improvement is not feasible.
Te variation of gradient descent using a rectilinear-regression model is
more commonly used as it is relatively straightforward to understand. Tis
algorithm fnds application in the scenario when the dataset is large and
hence won’t ft into the memory.

[Link] Applications of Linear Regression

Linear regression is a simple yet very sophisticated algorithm that fnds application
in a wide variety of felds. Roy et al. have proposed a Lasso Linear Regression Model
for stock-market forecasting [9]. Zameer et al. have used a linear-regression-based
model for predicting crude-oil consumption [10]. In general, linear-regression mod-
els are quite good in performing predictive data analytics.

2.4.2 Logistic Regression

Te technique of logistic regression in big data analytics is used when the variable
to be considered is dichotomous (binary). Te basis of logistic regression, just
like all other regression, is a predictive analysis. Logistic regression is employed
to elucidate data and to explain the connection between one dependent binary
variable and one or more nominal, ordinal, interval, or ratio-level independent
variables.
Logistic regression works on the concept of logit—the natural logarithms of an
odds ratio [11]. Tis type of regression model works quite well when the dependent
variable is categorical. Some examples of real-world problems where the dependent
variable can be categorical are predicting if the email is spam (1) or not (0) or if a
tumor is malignant (1) or safe (0). Logistic regression is a component of a bigger
class of algorithms referred to as the generalized linear model (GLM). In 1972,
 Big Data Analytics and Algorithms  31

Figure 2.6 A Sample Logistic-Regression Plot.

Nelder and Wedderburn proposed this model in an attempt to supply a way of

using rectilinear regression with the issues that weren’t directly ftted to the appli-
cation of rectilinear regression. Tey proposed a category of various models (linear
regression, ANOVA, Poisson regression, etc.), including logistic regression as a spe-
cial case. Equation 2–3 represents a general equation of logistic regression.

loglog {1− p} = ˜0 + ˜1 * x (2–3)

(p/1‑p) = odd ratio

x = independent variable
˜1 = scale factor
˜0 = bias coefcient

In this equation {1− p} is the odds ratio. Te positive log of an odds ratio usually
translates into a probability of success greater than 50%. A sample plot of logistic
regression is shown in Figure 2.6.

[Link] Types of Logistic Regression

1� Binary Logistic Regression
In binary logistic regression, a categorical response can only have two possible
outcomes. Example: Spam or Not email.
32  Big Data Analytics

2� Multinomial Logistic Regression

In multinomial logistic regression, dependent (target) variables can have three
or more categories without ordering. Example: predicting which food is pre-
ferred more (Veg, Non-Veg, Vegan).
3� Ordinal Logistic Regression
Ordinal logistic regression is a subset of multinomial logistic regression in
which dependent (target) variables can have three or more categories but in a
defned order. Example: movie rating from 1–5.

[Link] Applications of Logistic Regression

Logistic regression is a simple yet efcient algorithm that fnds application in a wide
variety of felds. Due to its predictive nature, logistic regression fnds application in
felds ranging from education to healthcare. Ramosaco et al. have developed a logistic-
regression-based model to study students’ performance levels [12]. Alzen et al. have
proposed another logistic-regression-based model to fnd the relationship between the
learning assistant model and failure rates in introductory STEM courses [13].
Although linear regression and logistic regression are both regression-based
models, they do share a lot of diferences. Tese diferences are shown in Table 2.2.

Table 2.2 Difference between Linear and Logistic Regression

Linear Regression Logistic Regression

Linear regression is used to predict Logistic regression is used to predict

the continuous dependent variable the categorical dependent variable
using a given set of independent using a given set of independent
variables. variables.

Linear regression is used for solving Logistic regression is used for solving
the regression problem. classifcation problems.

In linear regression, we predict the In logistic regression, we predict the

value of continuous variables. values of categorical variables.

In linear regression, we fnd the In logistic regression, we fnd the

best-ftting line, by which we can S-curve by which we can classify the
easily predict the output. samples.

The least-square estimation method The maximum-likelihood estimation

is used for the estimation of accuracy. method is used for the estimation of
accuracy.

The output of linear regression must The output of logistic regression

be a continuous value, such as price, must be a categorical value such as 0
age, etc. or 1, Yes or No, etc.
 Big Data Analytics and Algorithms  33

Linear Regression Logistic Regression

In linear regression, it is required that In logistic regression, it is not

the relationship between the required to have the linear
dependent variable and independent relationship between the dependent
variable be linear. and independent variable.

In linear regression, there may be In logistic regression, there should

collinearity between the independent not be collinearity between the
variables. independent variables.

2.4.3 Naive Bayes Classifers

Naive Bayes classifers are a set of classifcation algorithms supported by Bayes’
theorem. It’s not one algorithm but a family of algorithms where all of them share
a standard principle, i.e. every pair of features being classifed is independent of
every other.
Naive Bayes uses the probabilistic approach for constructing classifers. Tese
classifers can simplify learning by assuming that features are independent of given
class [14]. Naive Bayes classifcation is a subset of Bayesian decision theory. It’s
called naive because the formulation makes some naive assumptions [15].
Te main assumption that Naive Bayes classifers make is that the value of a
specifc feature is independent of the value of the other feature. Despite having an
oversimplifed assumption, Naive Bayes classifers tend to perform well even in
complex real-world scenarios. Te main advantage that Naive Bayes classifers have
over other classifcation algorithms is the requirement of a little amount of training
data for estimating the parameters necessary for classifcation, which is used for an
incremental training of the classifer.

[Link] Equation of the Naive Bayes Classifers

To understand the equation of Naive Bayes classifers we need to understand Bayes’
theorem, which is the fundamental theorem on which Naive Bayes classifers work.

Bayes’ theorem
Bayes’ theorem fnds the probability of the occurrence of an event,
given the probability of another event that has already occurred. Bayes
theorem is stated mathematically as shown in equation 2–4:

B 
P   * P ( A )
 A  A (2–4)
P   =
 B  P (B )
34  Big Data Analytics

P(A) = Probability of occurrence of event A

P(B) = Probability of occurrence of event B
P(A/B) = Probability of A given B
P(B/A) = Probability of B given A

Bayes’ theorem can be extended to fnd equations of various Naive

Bayes classifers.

[Link] Application of Naive Bayes Classifers

Naive Bayes classifers, despite having certain limitations and assumptions, work
quite well for solving classifcation problems. Karthika and Sairam propose a clas-
sifcation methodology utilizing the Naive Bayesian classifcation algorithm for
the classifcation of persons into diferent classes based on various attributes rep-
resenting their educational qualifcation [16]. Qin et al. research classifying mul-
tilabel data based on Naive Bayes classifers, which can be extended to multilabel
learning [17].

2.4.4 Classifcation and Regression Trees

Classifcation and regression trees (CART) is a term coined by Leo Breiman to
allude to the decision tree class of algorithms that are used to solve the classifcation
and regression predictive analytics problems.
Traditionally, this calculation is alluded to as ‘decision trees’; however, in cer-
tain programming languages like R they are alluded to by the more present-day
term CART. Te CART algorithms give an establishment for some other signif-
cant algorithms like bagged decision-tree algorithms, random-forest algorithms,
and boosted decision-tree algorithms.

[Link] Representation of CART Model

Te CART model can be represented as a binary tree. Each node in the tree repre-
sents a single input variable (x) and a split point theorem variable, and the leaf node
is represented using an output variable (y), which is utilized for forecasting.
For example, suppose a dataset having two input variables (x) of height in cen-
timeter and weight of a person in kilogram the output variable (y) will tell whether
the sex of the person is male or female. Figure 2.7 represents a very simple binary
decision tree model.
A straightforward way for making predictions using the CART model is with
the help of its binary tree representation. Te traversal of the tree starts with the
evaluation of a specifc input starting with the root node of the tree. Each input
variable in the CART model can be thought of as a dimension in an n-dimensional
space. Te decision tree in this model splits this plane into rectangles for two input
 Big Data Analytics and Algorithms  35

Figure 2.7 Representation of Binary Decision-Tree Model.

variables or into hyperrectangles for higher inputs. Te input data gets fltered
through the tree and gets placed in one of the rectangles, whereas the prediction
made by the model is the output value for the same rectangle; this gives us some
idea about the type of decisions that a CART model is capable of making, e.g. boxy
decision boundaries.

[Link] Application of Classifcation and Regression Trees

Pham et al. have used a classifcation and regression tree-based model for predicting
the rainfall-induced shallow landslides in the state of India based on a dataset of
430 historic landslide locations [18]. Pouliakis et al. have done a study on CART-
based models to estimate the risk for cervical intraepithelial neoplasia [19]. Iliev et
al. have proposed a CART-based model for modeling the laser output power of a
copper bromide vapor laser [20].

2.4.5 K-Means Clustering

K-means clustering is a very simple yet popular data-analytics algorithm. It is an
unsupervised algorithm as it capable of drawing conclusions from datasets having
only input variables without the requirement of having known or labeled outcomes.
Te goal of the K-means algorithm is very basic: just group similar data points and
reveal the pattern present in the dataset. K-means tries to fnd a predefned number
(k) of the cluster in the dataset. A cluster in very simple terms can be thought of
as a group of similar data points. Te prerequisite of the algorithm is the target
number k, which denotes the number of centroids required by us. A centroid can
either be a real or an imaginary point that represents the center of one single clus-
ter. Each information point is designated for every one of the groups by reducing
36  Big Data Analytics

the in-cluster sum of squares. Te K-means algorithm distinguishes the predefned

number of centroids and afterward allots each data point to the nearest cluster, with
the goal being to keep the centroids as tiny as could be expected. Te ‘means’ in the
K-means alludes to the aggregation of the information or, say, fnding the centroid.

[Link] How K-Means Clustering Works

For handling the learning information, the K-means algorithm in data analytics
begins with a set of randomly selected centroids; these are utilized as the starting
point for each cluster and afterward perform iterative calculations to improve the
places of the centroids.
It stops making and optimizing cluster when either of the conditions is met:

 Te centroids have stabilized and the algorithm can proceed further, i.e. the
clustering has been successful.
 Te predefned number of iterations has been reached.

[Link] The K-Means Clustering Algorithm

Te K-means clustering algorithm follows the approach of expectation-maximi-
zation. Te expectation step is assigning the data point to the closet cluster. Te
maximization step is fnding the centroid of each of these clusters. Te fnal goal of
the K-means algorithm is to minimize the value of squared error function given as:

c ci
2
J (V ) = ∑∑ ( x −v )
i=1 j=1
i j

xi −v j is the Euclidean distance between x and v

i j

[Link] Application of K-Means Clustering Algorithms

Being a high performing, unsupervised learning algorithm, K-means fnds applica-
tion in a wide variety of felds. Due to its popularity, researchers have created dif-
ferent hybrid versions of this algorithm that are being used extensively in numerous
felds. Youguo & Haiyan have developed a clustering algorithm on top of K-means
clustering, which provides greater dependence to choose the initial focal point [21].
Shakil and Alam have devised a method for data management in the cloud-based
environment on the basis of the K-means clustering algorithm [22]. Alam and
Kishwar have categorized various clustering techniques that have been applied to
web search results [23]. Alam and Kishwar have proposed an algorithm for web-
search clustering based on K-means and a heuristic search [24].
 Big Data Analytics and Algorithms  37

2.5 Conclusion and Future Scope

In this chapter, we looked into the basics of data analytics along with its applica-
tion in the real world. We also looked into various categories of data analytics along
with some of the most commonly used data-analytics algorithms as well as their
applications to the real-world scenario. Apart from the algorithms discussed in this
chapter, data scientists all over the world have been working on designing faster and
more efcient algorithms. Te idea of using neural-network-based algorithms has
been also proposed by data scientists [25, 32]. With the rise of quantum comput-
ing in the last couple of years, scientists are also looking forward to the possibility
of leveraging the power of quantum computers in big-data analytics [26]. Cloud-
based big-data analytics is also becoming quite popular as it can leverage the power
of cloud computing for big-data analytics [27–31]. With these new technological
advancements on the horizon, it can be safely assumed that the future of big-data
analytics is going to be bright and exciting.

References
1. Shakil, K. A., Anis, S., & Alam, M. (2015). Dengue disease prediction using weka
data mining tool. arXiv preprint arXiv:1502.05167.
2. Khan, M. W., & Alam, M. (2012). A survey of application: Genomics and genetic
programming, a new frontier. Genomics, 100(2), 65–71.
3. Shakil, K. A., Zareen, F. J., Alam, M., & Jabin, S. (2020). BAMHealthCloud: A
biometric authentication and data management system for healthcare data in cloud.
Journal of King Saud University‑Computer and Information Sciences, 32(1), 57–64.
4. Khanna, L., Singh, S. N., & Alam, M. (2016, August). Educational data mining and
its role in determining factors afecting students academic performance: A systematic
review. In 2016 1st India International Conference on Information Processing (IICIP)
(pp. 1–7). New York: IEEE.
5. Shakil, K. A., Sethi, S., & Alam, M. (2015, March). An efective framework for man-
aging university data using a cloud based environment. In 2015 2nd International
Conference on Computing for Sustainable Global Development (INDIACom) (pp. 1262–
1266). New York: IEEE.
6. Bagherian, M., Sabeti, E., Wang, K., Sartor, M. A., Nikolovska-Coleska, Z., &
Najarian, Z. (2021). Machine learning approaches and databases for prediction of
drug—target interaction: a survey paper. Briefngs in Bioinformatics, 22(1), 247–269.
[Link]
7. Šikšnys, L., Pedersen, T. B., Liu, L., & Özsu, M. (2016). Prescriptive analytics.
Encyclopedia of Database Systems, 1–2.
8. Kaya Uyanık, G., & Güler, N. (2013). A study on multiple linear regression analysis.
Procedia—Social and Behavioral Sciences, 106, 234–240. [Link]
sbspro.2013.12.027.
9. Roy, S. S., Mittal, D., Basu, A., & Abraham, A. (2015). Stock market forecast-
ing using LASSO linear regression model. In: Abraham, A., Krömer, P., &
38  Big Data Analytics

Snasel, V. (eds.), Afro‑European Conference for Industrial Advancement. Advances

in Intelligent Systems and Computing, vol. 334. Cham: Springer. [Link]
org/10.1007/978-3-319-13572-4_31.
10. Fatima, Z., Kumar, A., Bhargava, L., & Saxena, A. (2019). Crude oil consumption
forecasting using classical and machine learning methods. International Journal of
Knowledge Based Computer Systems, 7(1), 10–18.
11. Peng, J., Lee, K., & Ingersoll, G. (2002). An introduction to logistic regres-
sion analysis and reporting. Journal of Educational Research, 96, 3–14. [Link]
org/10.1080/00220670209598786.
12. Ramosaco, M., Hasani, V., & Dumi, A. (2015). Application of logistic regression in
the study of students’ performance level (Case Study of Vlora University). Journal of
Educational and Social Research, 5(3). [Link]
13. Alzen, J. L., Langdon, L. S., & Otero, V. K. (2018). A logistic regression investigation
of the relationship between the Learning Assistant model and failure rates in intro-
ductory STEM courses. International Journal of STEM Education, 5, Article number
56. [Link]
14. Rish, I. (2001). An empirical study of the naïve bayes classifer. IJCAI 2001 Work
Empir Methods Artif Intell, 3.
15. Kaviani, P., & Dhotre, S. (2017). Short survey on naive bayes algorithm. International
Journal of Advance Research in Computer Science and Management, 4.
16. Karthika, S., & Sairam, N. (2015). A naïve bayesian classifer for educational qualif-
cation. Indian Journal of Science and Technology, 8. [Link]
v8i16/62055.
17. Qin, F., Tang, X., & Cheng, Z. (2012). Application and research of multi_label
Naïve Bayes Classifer. In Proceedings of the 10th World Congress on Intelligent
Control and Automation (pp. 764–768). New York: IEEE. [Link]
WCICA.2012.6357980.
18. Pham, B. T., Tien Bui, D., & Prakash, I. (2018). Application of classifcation and
regression trees for spatial prediction of rainfall-induced shallow landslides in the
Uttarakhand Area (India) using GIS. In: Mal, S., Singh, R., & Huggel, C. (eds.),
Climate Change, Extreme Events and Disaster Risk Reduction. Sustainable Development
Goals Series. Cham: Springer. [Link]
19. Pouliakis, A., Karakitsou, E., Chrelias, C., Pappas, A., Panayiotides, I., Valasoulis,
G., Kyrgiou, M., Paraskevaidis, E., Karakitsos, P. (2015). Te application of classif-
cation and regression trees for the triage of women for referral to colposcopy and the
estimation of risk for cervical intraepithelial neoplasia: A study based on 1625 cases
with incomplete data from molecular tests. BioMed Research International, 2015,
Article ID 914740, 10 p. [Link]
20. Iliev, I. P., Voynikova, D. S., & Gocheva-Ilieva, S. G. (2013). Application of the
classifcation and regression trees for modeling the laser output power of a copper
bromide vapor laser. Mathematical Problems in Engineering, 2013, Article ID 654845,
10 p. [Link]
21. Li, Y., & Wu, H. (2012). A clustering method based on K-means algorithm. Physics
Procedia, 25, 1104–1109. [Link]
22. Shakil, K. A., & Alam, M. (2014). Data management in cloud based environment
using k-median clustering technique. International Journal of Computer Applications,
3, 8–13.
 Big Data Analytics and Algorithms  39

23. Alam, M., & Sadaf, K. (2013). A review on clustering of web search result. In:
Advances in Computing and Information Technology (pp. 153–159). Berlin, Heidelberg:
Springer.
24. Alam, M., & Sadaf, K. (2015). Web search result clustering based on heuristic search
and K-means. arXiv preprint arXiv:1508.02552.
25. Mamatha, C., Reddy, P., Kumar, M. A., & Kumar, S. (2017). Analysis of big data
with neural network. International Journal of Civil Engineering and Technology, 8,
211–215.
26. Shaikh, T. (2016). Quantum computing in big data analytics: A survey. Conference:
2016 IEEE International Conference on Computer and Information Technology (CIT)
(pp. 112–115). [Link]
27. Khan, S., Shakil, K. A., & Alam, M. (2017). Cloud based big data analytics: A survey
of current research and future directions. Big Data Analytics, Print ISBN: 978-981-
10-6619-1, Electronic ISBN: 978-981-10-6620-7, (pp. 629–640). Springer.
28. Alam, M. (2012). Cloud algebra for cloud database management system. Te Second
International Conference on Computational Science, Engineering and Information
Technology (CCSEIT‑2012), October 26–28, Coimbatore, India, Proceeding pub-
lished by ACM.
29. Alam, M. (2012). Cloud algebra for handling unstructured data in cloud data-
base management system. International Journal on Cloud Computing: Services and
Architecture (IJCCSA), 2(6), ISSN: 2231–5853 [Online]; 2231–6663 [Print]. https://
[Link]/10.5121/ijccsa.2012.2603, Taiwan.
30. Alam, M., & Shakil, K. (2013). Cloud database management system architecture.
UACEE International Journal of Computer Science and its Applications([ISSN 2250–
3765), 3, 27–31.
31. Alam, B., Doja, M. N., Alam, M., & Malhotra, S. (2013). 5-layered architecture
of cloud database management system. AASRI Procedia Journal, 5, 194–199, ISSN:
2212–6716, Elsevier.
32. Alam, M., Shakil, K. A., Mohd. Javed, S., & Ambreen, M. A. (2014). Detect and fl-
ter trafc attack through cloud trace back and neural network. Te 2014 International
Conference of Data Mining and Knowledge Engineering (ICDMKE), Imperial College,
London, UK, 2–4 July. Hong Kong: IAENG.