See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/351087376

Machine Learning and Deep Learning: A Comparative Review

Chapter · April 2021

DOI: 10.1007/978-981-33-6307-6_15

CITATIONS READS
23 4,622

2 authors:

Hind Alaskar Tanzila Saba

Prince Sultan University Prince Sultan University
7 PUBLICATIONS 127 CITATIONS 490 PUBLICATIONS 20,719 CITATIONS

SEE PROFILE SEE PROFILE

All content following this page was uploaded by Hind Alaskar on 10 June 2021.

The user has requested enhancement of the downloaded file.

 Machine Learning and Deep Learning: A
Comparative Review?

Hind Alaskar1 and Tanzila Saba2

1
Prince Sultan University, College of Computer and Information Sciences, AIDA
Lab, Riyadh, Saudi Arabia
2
Prince Sultan University, College of Computer and Information Sciences, AIDA
Lab, Riyadh, Saudi Arabia
[email protected] [email protected]

Abstract. Machine Learning and Deep Learning are revolutionary fields

in the computer and information sciences area and are used widely in
business applications. Machine Learning is an approach to train comput-
ers or machines to learn from past data or behaviour so it can determine
future data or behaviour. Deep Learning is a branch of Machine Learning
where the Artificial Neural Network techniques and algorithms are used
to train and learn. Due to the large volume of data points generated
daily, there is a great need for the proper use of data and data man-
agement using the major techniques. This paper offers a comparative
review of Machine Learning (ML) and Deep Learning (DL). The paper
starts by providing an introduction to the definition of ML and DL.
Then, the main approaches and techniques used to enable the technolo-
gies are discussed. At the end, the paper presents two popular examples
of applications used by ML and DL from the business perspective.

Keywords: Machine Learning · Deep Learning, · Artificial Intelligence

1 Introduction
Artificial Intelligence is a technique that enables
computers or machines to mimic human ability
to perform, behave, or function [1]. AI can learn
from past data and is able to recognize patterns
and behavior. It makes it possible for a machine
to learn from experience and tolerate its behav-
ior. The term AI was coined in 1956 by John
McCarthy [2].There are many interdisciplinary
fields under AI. Fig. 1. represents the position
of ML and DL in relation to AI [3]. The term
AI is very broad while Machine Learning (ML)
is a branch of AI [4]. ML was firstly coined in Fig. 1: Example of AI and ML
?
Supported by Artificial Intelligence and Data Analytic Lab, Prince Sultan University,
.
2 Alaskar H, Saba T

1959 by Arthur Samuel which uses statistical

methods to enable machines to improve their
performance and behavior through experience [5]. According to [3], Deep Learn-
ing is a sub-branch of Machine Learning. The term was introduced in 1986 to
Machine Learning community by Rina Dechter and in 2000 to Artificial Neural
Network by Igor Aizenberg [6]. The paper will start with an overview of ML and
DL, then it provides the major details on the approaches and techniques used
to enable ML and DL. The last section presents two major applications used in
daily life which are supported by ML and DL.

2 Basic Concepts: An Overview

2.1 Machine Learning

Machine Learning (ML) is not a new term in the computer science field. However,
it is a very growing and evolving term. ML is a subject that study how to use
machines to replicate human’s capability and behavior. It is considered as a
technology that is used to answer questions and derives meanings from data. It
brings together statistics and computer science to enable computers or machines
to perform a given task. Unlike humans, ML has the capability to learn and adapt
faster [7]. It is the subset of Artificial Intelligence where techniques, models, and
algorithms are used to learn from data and information to derive knowledge,
patterns, or deliver AI applications. ML consists of set of algorithms that are
used as the method to train historical data set and make future predictions. Due
to the rise of technology and internet, data is generated rapidly - which means
there are huge amount of data for the machine to learn and analyze [4]. There are
three main categories under ML - Supervised Learning, Unsupervised Learning,
and Reinforcement Learning [7]. These categories are discussed in section 3 -
Main Approaches and Techniques.

2.2 Deep Learning

Deep Learning is a very popular and widely used subset of Machine Learning. It
combines Artificial Intelligence and Machine Learning together [6]. Deep Learn-
ing relies on multi-layered Artificial Neural Network (ANN) [8]. The concept
is based on the biological neural networks in humans. These networks contain
nodes and multi-layers that communicate with each other to understand and
analyze the data received. DL consists of several layers, while each layer may
consist of many neurons. The neuron in each layer is called Processing Unit. The
main role of neurons is to multiply the input value with the given weight and
calculate the sum of the results [9]. This is how Deep Learning models acquire
higher accuracy. Deep Learning architecture is discussed in section 3 - Main
Approaches and Techniques.
 Machine Learning and Deep Learning: A Comparative Review 3

3 Main Approaches and Techniques

3.1 Machine Learning
Machine Learning includes set of well-defined computer instructions that are
followed and performed by computers or machines to solve a specific problem
or perform a specific action [10]. ML is used to train set of data, learn, adjust,
and make the prediction or action. One simple example of machine prediction is
being able to find probability of loan payment or terrorist suspect.
There are three important techniques used in Machine Learning concept
which vary according to the type of ML problem and the input data. These
techniques are Supervised Learning, Unsupervised Learning, and Reinforcement
Learning. [11].Table 1 summarizes these techniques along with its most popular
learning algorithms.

Table 1: Summary of Machine Learning Techniques

Techniques Supervised Learning Unsupervised Learning Reinforcement Learning
Classification Clustering
Methods Control Learning
Regression Association
Value Based
Linear Classifier
Hierarchical clustering Policy Based
Logistic Regression
k-means Model Based
Support Vector Machine
Mixture models State Value Function
Naive Bayesian Network
Algorithms DBSCAN Brute Force
K-Means
OPTICS algorithm Value Functions
Decision Tree
Apriori Algorithm Monte Carlo
Neural Network
FT Growth Algorithm Temporal Difference
Multilayer Perceptron
Direct Policy Search
References [12] [13][14] [15]

Supervised Learning Supervised Learning is one of the most popular tech-

nique of ML that learn from experience based on historical data set and perfor-
mance measure. Supervised Learning depends on mapping the input data to the
target output by learning from historical data set the best way to map the in-
put to target output. Supervised Learning deals with labelled data and produce
labelled data as well.[12]
Supervised Learning algorithms study the labelled data. Thus, it would be
able to determine the label should be given to new unlabelled data based on
learning and observation. The main methods used in Supervised Learning are
Classification and Regression. Both methods are used to build prediction models
based on data. However, the target prediction differs in terms of technique used
in ML. In classification technique, the target prediction (output) is categorical
or discrete. It predicts a categorical item from a list it belongs to. On the other
4 Alaskar H, Saba T

hand, regression technique produces numerical or continues output data. For

example, it predicts human height, or house price. Based on the type of desired
target prediction (output), certain algorithms are used. The explanation of each
algorithms is out of the scope of this paper. The most popular algorithms in each
Supervised Learning techniques- Classification or Regression, are Linear Classi-
fier, Logistic Regression, Support Vector Machines (SVM), Naive Bayesian (NB)
Networks, K-Means, Decision Tree, Neural Network, and Multi-layer Perceptron.
[12]

Unsupervised Learning The second technique of ML is Unsupervised Learn-

ing. It is used to build a prediction model when there are no preexisting data
label to train on. Unsupervised Learning depends on the discovery of patterns
in the data as it handles unlabelled data.
Unsupervised Learning algorithms study the unlabelled data and produce
new groups with related data. The algorithm role in Unsupervised Learning is
to define the hidden structure in unlabelled data/ or lass. It accomplishes this
by providing the learning algorithms with large number of input data, charac-
teristics, and features so the algorithms can make necessary observation to find
out the best grouping or relationships between data [14].
The main methods used in unsupervised learning are clustering and associ-
ation. Clustering is the process of aligning each data entry with groups (called
cluster) in such a way that each data entry in the group share similar charac-
teristics or feature such as shape, colour, or size. The most popular algorithms
for clustering are K-means, and Hierarchical clustering. On the other hand, As-
sociation method is used to find the dependency between one data entry to
another and map them together. The most popular algorithms for association
are Apriopri Algorithm and FT Growth Algorithm.[13]

Reinforcement Learning The third technique of ML is Reinforcement Learn-

ing. RL uses past data with no predefined label to improve the future manipula-
tion of dynamic system. [16] It is a learning technique where an agent interacts
with an environment by producing actions and discover errors or rewards. The
agent is put in an unknown environment to behave and learn by receiving sig-
nals. These signals can be rewards or punishment depending on the agent action.
It allows the agent to adjust action and learn better for future actions. [15]
RL aim to maximize the reward signals rather than finding patterns or la-
belling data as the case in Supervised and Unsupervised Learning. In RL, the
communication between the agent and its environment is stated in a formal
framework in terms of state, action, signal (reward or punishment) [16]. The
framework represents the typical representation of artificial intelligence. The
most popular algorithms and techniques for RL are Monte-Carlo, Temporal Dif-
ference Methods, SARSA (State–Action–Reward–State–Action), and Direct Pol-
icy Search, Q-learning) [15].
 Machine Learning and Deep Learning: A Comparative Review 5

3.2 Deep Learning

Deep Learning share similar characteristics with Machine Learning. Both con-
cepts rely on training models and classifying data. DL offers the computer the
ability to learn from and recognize examples. DL is a subset of Machine Learn-
ing but more complex. DL need to use big data along with high computing
performance to deal with the data [17]. It deals with Image Processing, Natural
Language Processing (NLP), or Voice Recognition. Deep Learning uses differ-
ent approaches to train data depending on the type of data. The most popular
approaches are Convolution Neural Network (CNN), Recurrent Neural Network
(RNN), Long Short Term Memory (LSTM), and Generative Adversarial Net-
works (GAN)[18].
CNN Approach has the power to deal with images by extracting different
features[18]. According to [19], the CNN architecture represents the four different
layers. The input layer holds the image data. The second layer, convolution layer,
is the core block of CNN. It gets activation features in the image by applying
several filters on the input image. In addition, Pooling layer aim to downsize
sample in order to reduce memory and computation issues. The next layer is the
Fully-Connected layer which aim to detect and identify the final output category.
Each node/input is connected to a coefficient which makes this layer heavy.

4 Main Applications

4.1 Machine Learning

Virtual Personal Assistants VPAs are intelligent systems that allow users
to interact with virtual agents to finish some tasks. VPAs are incorporated into
different devices such as Smart TVs, Smart-Phones, Applications, and Car Nav-
igation Systems. [20] There are several well-known companies who used VPA
technology supported by ML in their systems such as Microsoft Cortona, Ap-
ple Siri, and Amazon Alexa. Companies implemented different techniques and
algorithms to develop and enhance their applications. Microsoft Cortana is a
ML application developed by Microsoft to answer users questions and perform
their tasks. It can set reminder, answer user’s questions using Bing search en-
gine, and recognize natural voices to form sentences. Cortana application uses
ML algorithms to be able to perform users’ tasks efficiently such as predicting
when to leave the house to avoid traffics. Similar applications are developed by
other companies who share the same functionality and Microsoft Cortana. Most
companies are enhancing their applications to make their VPA software better.
[21]

Email Spam Detection Spamming (as known as Spam Email) is the act of
sharing something in the form of messages among number of recipients. However,
Ham Emails are normal Emails sent by legitimate user. Most spam email are
directed by spammers for marketing purpose and are annoying to most Email
6 Alaskar H, Saba T

users.[22] There are many companies who developed spam detection software
supported by ML. Machine Learning has shown great support to the enhance-
ment of those SW due to its ability to learn and tune itself by time. Spam Email
detection uses different ML algorithms and methods (supervised, unsupervised,
or reinforcement). Some of the most used algorithms in Email spam detections
are ANN, Naı̈ve Bayes, Decision Tree, Logistic Regression, and SVMs. Overtime,
more algorithms are being used and more models are being enhanced to detect
better.[23]

4.2 Deep Learning

Recommendation Systems recommender systems aim to predict and suggest
user preference based on history or some related items. For example, it predicts
related products based on user past purchases. It is one of the most widely used
systems in e-commerce websites. These systems are supported by ML algorithms
to predict with high accuracy/precision. Deep learning algorithms have moved
the system into another level in terms of the way it works. DL algorithms allow
analysing input images and recommending related images. This process is known
as Content-Based Image Retrieval (CBIR); which takes input image by user and
search in large database for similar/related images. CBIR uses techniques to
decompose input images into components. It is used widely in different industries
such as business, and healthcare industry.[24]

Face Recognition Systems FRS is one of the most popular DL applications.

It aims to recognize people faces by their face characteristic and provide identi-
fication. It is a challenging computer vision application since people faces may
change by age, surgery, or when applying cosmetic makeup. DL models has ef-
ficiently shown a great success in this field and exceeded human expectations.
CNNs are one of the most popular approaches used in DL recognition systems.
Most DL algorithms aim to perform few main steps to accomplish identification
objective. First, it detects the face in the image and form the bounding with
boxes. Next, it normalizes the face in terms of photo-metrics and geometry to
be aligned with the database. Then, it extracts important features from the face
that can be used in identification and recognition. Finally, it matches the face
with similar pictures sharing similar features to provide identification.[25]

5 Conclusion
Artificial Intelligence enhanced the way computers and machines work. It in-
cludes several branches. However, the paper focused on providing a brief review
of machine learning and deep learning. These concepts are highly dynamic and
evolving overtime. Authors and AI practitioner have different views and perspec-
tives on how these technologies work. Data and Intelligence is the common base
between them. For future research it is suggested to focus on one of the concepts
and study the process of learning algorithms used to enable the technology.
 Machine Learning and Deep Learning: A Comparative Review 7

References
1. Mitchell M (2019) Artificial Intelligence: A Guide for Thinking Humans. Farrar,
Straus and Giroux
2. Wilks Y (2019) Artificial Intelligence: Modern Magic or Dangerous Future. Icon
Books
3. Campesato O (2020) Artificial Intelligence, Machine Learning, and Deep Learning.
Mercury Learning and Information
4. Joshi, Ameet (2019) Introduction to AI and ML. 10.1007/978-3-030-26622-6-1
5. Bell J (2020) Machine Learning: Hands-On for Developers and Technical Profes-
sionals. Wiley
6. Agrawal A, Agrawal R, and Gupta A (2018) Extracting Knowledge from Opinion
Mining. IGI Global, USA
7. Berry, Michael W., Mohamed, Azlinah H, Yap, Bee Wah (2019) Supervised and
Unsupervised Learning for Data Science. Springer Nature
8. Modi A S (2018) Review Article on Deep Learning Approaches, Second International
Conference on Intelligent Computing and Control Systems (ICICCS), Madurai, In-
dia, pp. 1635-1639, doi: 10.1109/ICCONS.2018.8663057.
9. Graupe D (2019) Principles Of Artificial Neural Networks: Basic Designs To Deep
Learning. Fourth Edition, World Scientific
10. Sethi, Kapil Gupta, Ankit Gupta, Gaurav Jaiswal, Varun (2019) Comparative
Analysis of Machine Learning Algorithms on Different Datasets
11. Ray S (2019) A Quick Review of Machine Learning Algorithms, International Con-
ference on Machine Learning, Big Data, Cloud and Parallel Computing (COMIT-
Con), Faridabad, India
12. Bzdok D, Krzywinski M Altman N. (2018) Machine learning: supervised methods.
Nat Methods 15, 5–6 https://doi.org/10.1038/nmeth.4551
13. Dike H U, Y, Zhou, Deveerasetty K and Wu Q (2018) Unsupervised Learning Based
On Artificial Neural Network: A Review, IEEE International Conference on Cyborg
and Bionic Systems (CBS), Shenzhen, pp. 322-327, doi: 10.1109/CBS.2018.8612259
14. Solorio-Fernández, S., Carrasco-Ochoa, J.A. and Martı́nez-Trinidad, J.F. (2020)
A review of unsupervised feature selection methods. Artificial Intelligence Review,
53(2), pp.907-948
15. Benjamin R (2018) A Tour of Reinforcement Learning: The View from Continuous
Control. CoRR abs/1806.09460. arXiv:1806.09460 http://arxiv.org/abs/1806.09460
16. Zhang, Han X and Deng C, (2018) Review on the research and practice of deep
learning and reinforcement learning in smart grids, CSEE Journal of Power and
Energy Systems, vol. 4, no. 3, pp. 362-370, doi: 10.17775/CSEEJPES.2018.00520
17. Eugene C (2019) Introduction to Deep Learning. The MIT Press.
18. Shrestha A and Mahmood A (2019) Review of Deep Learning Algorithms
and Architectures, in IEEE Access, vol. 7, pp. 53040-53065, doi: 10.1109/AC-
CESS.2019.2912200
19. Pouyanfar, S., Sadiq, S., Yan, Y., Tian, H., Tao, Y., Reyes, M.P., Shyu, M.L., Chen,
S.C. and Iyengar, S.S. (2018) A survey on deep learning: Algorithms, techniques,
and applications. ACM Computing Surveys (CSUR), 51(5), pp.1-36
20. Tulshan, A.S. and Dhage, S.N. (2018) Survey on Virtual Assistant: Google Assis-
tant, Siri, Cortana, Alexa. In International Symposium on Signal Processing and
Intelligent Recognition Systems (pp. 190-201). Springer, Singapore
21. Këpuska and Bohouta G (2018) Next-generation of virtual personal assistants (Mi-
crosoft Cortana, Apple Siri, Amazon Alexa and Google Home, IEEE 8th Annual
 8 Alaskar H, Saba T

Computing and Communication Workshop and Conference (CCWC), Las Vegas,

NV, pp. 99-103, doi: 10.1109/CCWC.2018.8301638
22. Dada, E.G., Bassi, J.S., Chiroma, H., Adetunmbi, A.O. and Ajibuwa, O.E. (2019)
Machine learning for email spam filtering: review, approaches and open research
problems. Heliyon, 5(6), p.e01802.
23. Karim, Azam S, Shanmugam B, Kannoorpatti K and Alazab M (2019) A Com-
prehensive Survey for Intelligent Spam Email Detection, in IEEE Access, vol. 7, pp.
168261-168295, doi: 10.1109/ACCESS.2019.2954791.
24. Zhang, S., Yao, L., Sun, A. and Tay, Y. (2019) Deep learning based recommender
system: A survey and new perspectives. ACM Computing Surveys (CSUR), 52(1),
pp.1-38
25. Masi I, Wu Y, Hassner T and Natarajan P (2018) Deep Face Recognition: A Sur-
vey. 31st SIBGRAPI Conference on Graphics, Patterns and Images (SIBGRAPI),
Parana, 2018, pp. 471-478, doi: 10.1109/SIBGRAPI.2018.00067.

View publication stats