Machine

Learning
d-25
Light="red"
o/wn

d=9' 6 d-9 10"

O=.5" O=.6

on

Anuradha Srinivasaraghavan
Vincy Joseph

WILEYY
Machine
Learning

Anuradha Srinivasaraghavan
Associate Professor,
Department of Computer Engineering,
St. Francis Institute
of Technology,
Mumbai, Maharashtra

Vincy Joseph
Assistant Professor,
Department of Computer Engineering,
St. Francis Institute
of Technology,
Mumbai, Maharashtra

WILEY
Contents

Preface vii
About the Authors
ix

Part 1 Basics of Machine

Learning
Chapter 1 Introduction to Machine Learning 3
Learning Objectives
Learning Outcomes
1.1 What is Machine Learning?
1.1.1 Common Definitions of Machine
Where is Machine Learning Used?
Learningx
1.2 6
13 Applications of Machine Learning
1.3.1 Marketing and Sales
1.3.2 Search Engines
1.3.3 Transportation
14 Types of Machine Learning
1.4.1 Supervised Learning
8
1.4.2 Unsupervised Learning
10
1.4.3 Reinforcement Learning
Case Srudy 1: 12
Diagnosing Crop Disease with Machine Learning 14
Case Study 2: Learning to Decode the Immune System to Diagnose Disease 14
Summary
15
Multiple-Choice Questions 15
Very Short Answer Questions
Short Answer Questions 16
Review Questions 16
Answers 16
16

Chapter 2 Model and Cost

Function 17
Learning Objectives
Learning
2.1
Outcomes 17
Introduction 17
2.2 17
2.3 Representation of a Model
17
2.4
Ost Function Notation for Measuring the Accuracy of a Hypothesis Function
18
Measuring Accuracy of aHypothesis Function
21
 Introduction to Machine
Learning
LEARNING OBJECTIVES
.To make students understand the importance To understand how machine learning tech-
and significance of machinelearning niques can be used for practical applications.
To differentiate between different types of
machine learning techniques.

LEARNING OUTCOMES
Students will be able to define the term Students will be able to sketch appropri-
machine learning and comprehend different ate machine learning techniques for different
types of learning. applications.
Students will be able to differentiate berween
supervised, unsupervised and reinforcement
learning techniques.

1.1 What is Machine Learning?

Machine learning (MIL) is in computer science which involves
an area
teaching computers to do things
haturally by learning through experience. This means that the computer system is turning data into infor
matton. Machine learning algorithms derive information from data without the help of any model. The
m s adaptively improve their performance as we increase for example, the number of tuples in the
dataset. It is
equivalent to the fact that learning happens with experience.
lom M Mitchell, one of the patrons of machine learning, defined it as follows:
opuler program is said to learn from experience (E) with respect to some class oftasks (T) and performance
e(P, ifits performance at tasks in T, as measured by R improves with experience E.
Xample, a computer program that learns to play tic-tac-toe might improve its performance, as
k d by is ability to win the game at the class of tasks involved in playing the game (all X's and O's
itltolow ahorizontal, vertical or diagonal line), through experience obtained by playing games against
general, to have a wel-defined learning problem, we must identity the following threefeatures,
1. Class of tasks.
2.
3. Performance measurement that needs
Source of
to be improved
ved.

experience.
 4
CHAPTER 1/INTRODUCTTION TO
MACHINE
robot navigation in a maze. Here
LEARNIN
Consider as an example,
. Class of task: Reaching the end of the maze.
2. Performance measurement: Time taken to reach the end of the maze
3. Source of experience: Navigating the maze from start to finish by the robot.

1.1.1 Common Definitions of Machine Learning

Nvidia defines machine learning as "Machine learning at its most basic is the
practise of us
to parse data. learn from it, and then make a determination or prediction about somethino ingori
Stanford University defines it as "Machine learning is the science of getting computers to a world
being explicitly programmed.
McKinsey & Company states that "Machine learning is based on algorithms that can learnf
without relying on rules-based programming.
According to Pedro Domingos from Department of Computer Engineering University of Washiner
Machine states that "Machine
in
his paper A Useful
Few to Know
Things about
can figure out how to perform important tasks by generalizing from examples."
Learning algorit learning
Carnegie Mellon University's machine learning department has many research projects on the subje
According to the researches in that prominent university, "The field of machine learning seeks toanswerth
question How can we build computer systems that automatically improve with experience, and the fund
mental laws that govern all learning processes?
Machine learning lies at the intersection of computer science, engineering and statistics and ofie
appears in other disciplines as shown in Fig. 1.1. t can be applied to many fields-from politics to geo
ences.
Any field that needs to interpret and act on data can benefit from machine learning techniques
shown in the following examples.

Example 1.1
On online traffic networks, mobile applications estimate price of ride and time of arrival at destinaton
riae based
using machine learning. For instance, Uber ATG uses machine learning to define price of the b
on the demand of riders and trafic in that
region.

Example 1.2 orkplace,

Facebook, notices the friends in the list, profiles often visited, yourinterests,
groups you
whichin continuously
and
suggestion
are so on. Based on the information retrieved, Facebook gives you t

Computer
Science Statistics

Machine
Learning/

Engineering

Figure 1.1 Machine learning is an intersection.

 IS MACHINE LEARNING? 5
1.1 WHAT

Example 1.3
Cancidet that you purchased an item from Amazon. After your purchase, you may find pop-up adver-
icements for shopping The product recommendations may be related to what you purchased. For
ctance, if you purchased a mobile phone online, then the site from where you purchased it immedi-
aTely recommends a cover for the phone purchased as shown in Fig. 1.2.

ORDER PLACED TOTAL SHIP TO

27 December 2017
ORDER # 403-3322867-939
258.00
Order Details Invoice

Xiaomi Redmi MI 4 Defender Stylish Hard Back Armor Shock

Proof Case Cover with Back Stand Feature Black Return or replace items
(2017 Model)
Sold by: Online Retailing
198.00 Write a product review

ORDER PLACED TOTAL SHIP TO

ORDER#407-8260808-97
21 December 2017 7 10,398.00 Vincy Joseph Order Details Invoice

Redmi 4 (Black, 64GB)

Sold by: GreenMobiles
Write a product review
9,999.00

Figure 1.2 Recommendation of cover with purchase of phone.

Machine learning uses algorithms to find patterns in data and then model that
terns to make
uses a
recognizes those pat-
predictions on new data. This is nothing but pattern recognition (Fig. 1.3).
Machine learning can be implemented in the healthcare sector, providing opportunities
to facilitate
and enhance the work of medical
experts and to improve the efficiency and quality of medical care. In
medical diagnosis, the main interest is in establishing the existence of a disease followed by its accurate
Identihcation. Machine learning can improve the accuracy of medical diagnosis by analyzing historical
data of patients for various categories of diseases. Measurements in this application are typically results of
certain medical tests
(such as blood pressure, temperature and various blood tests), medical diagnostics
uch medical
as
images), presencelabsence/intensity ofvarious symptoms and basic physical information

Predictions

Training Model:
algorithm:
Finds
Recognizes
patterns
patterns
Data

New Data

Figure 1.3 Pattern recognition using machine learning.

 CHAPTER 1/INTRODUCTION TO MACHINE LEAD
ARNIN
results of these meac..
and so on). On the basiS of
the
measurements, a mod
weight,
sex,
predicts the chance
ofthe patient (age, that analyses the
test results
of
a new patient and
be created tests can be suggested. Proper prevene.
can model, further
the results of the
the disease. Based on

m e a s u r e s can
then be advised by doctors.
prognostic

Used?
Where is Machine Learning
1.2
like? Have you wondered how
Are you fascinated by how
Netfix recommends movies you might Goog
Machine learning is behind these technological advances. It represe
shows you such accurate search results?
evolution in the fhields of computer science,
data analysis, software engineering and artificial intellioen
a key
1.4 shows how machine learning is used in Google.
Figure

Jobs
Réear Miumba: 3taharastitra

Computer and IT Past 3 dayS Ful-tme Suence and Engimeering Management

Data Scuencel Machine Leaning Engineer- Mumbai (4-5 Years 0f.

Conhdential
Mumbai, lAaharashtra
via Analytics Vicnya

Over 1 month ago Fu-tme P

Artficial Intelligence / Machine learning

E ELITE RECRUITMENTS
T4umbai 1Maharashtra
ViE TImesJoDs

95 cays ago Ful-turme

Data Scieces-
froctab Fractal Analytics
1Aumbai tAaharashtra
Vid LinKedln

3oars ago Fu-tiiE

Figure 1.4 Google using machine learning

Last decade has
seen immense growth in the field of
"Indeed"

an American worldwide artificial intellig

ligence and data
SCience.

employment-related search engine for job listings. lt was lau

in November

004. As Single-topic search

a

in over 60 countries and 28 engine,

it is also an
example of vertical search. Inaceu on to
become

languages.
the highest-traffic job website in the In October 2010,
Indeed.com passed Monster.com
010, Indeed.com been a 2000

ncrease in data scientist" job searchesUnited

anda
States. According to their statistics,
te
has
tical plot
is given
i
Fig. 1.5. 50% increase in job listings. The
staud
sBunsod qof Bujyoeu jo efeue31ed
 8 CHAPTER 1/INTRODUCTION TO MACHINE LEARNIN
are
increasingly using machine learning technology to carry out data analysis and cdata
modeling to mak
informed decisions and their custormers
case
experience.

1.4 Types of Machine Learning

Two of the widely adopted machine learning methods are supervised learning and unsupervisec
most

learning. Supervised learning trains algorithms based on example input and output data that is labeled b»
humans. Unsupervised learning provides the
structure or pattern within its
algorithm with no labeled data in order to allow it to fine
input data. Let's explore these two methods in more detail.
1.4.1 Supervised Learning
In supervised learning.
the conmputer is
provided with example inputs that are labeled with their desirec
outputs. The purpose of this method is for the algorithm to be able to "learn"
output with the "taught" ourputs to find errors, and modify the model
by comparing its actua
therefore uses patterns to predict label values on additional unlabeled accordingly. Supervised learning
data. t is similar to how a teache=
teaches his/her students. The teacher, or tutor, examines the
the solution known him/her
performance of the students and corrects witl-
to desired by him/her.
or
For example, suppose you have a fruit basket and your task is to
arrange the fruit by type (Fig. l.6).

Figure 1.6 Fruit basket.

You can group the truits based on

any physical character. For example, if you arrange them
tocolor, then the groups will be
arranged as such: red color group-apples and cherries, according-
group-watermelons and grapes (Fig. 1.7). and green colo
The ocher feature is the size
of the fruit. We all know that apples and cherries are both round in
size of an
apple is bigger than that of a cherry. ln the second shape. But the-
group, watermelon is much bigger than in
grapes sie
Red color group: Apples and Cherries
Green color group: Watermelons and
Grapes

Figure 1.7 Fruit basket

categorization based on color.
1.4 TYPES OF MACHINE LEARNING .9

So if we want to separate the fruits from a fruit basket based on color, which is the first feature, then apples
and cherries would be categorized in one group and watermelon and grapes would be in another group.
But if we want to automate this process of fruit categorization then the feature "color" alone would not be
sufhcient to group them. So we make use of the additional feature "size". If the color is red and size is small
the fruit is classified as cherry ese apple, as shown in Fig. 1.8. In the case of green fruits, if the color is green
and size is big then the fruit is classifhed as watermelon else grape.

Red color and big size group: Apples Green color and big size group: Watermelons

Red color and small size group: Cherries Green color and small size group: Grapes

Figure 1.8 Fruit basket categorization based on color and size.

This brings us rules for automatic

to some
categorization of some fruits in the fruit basket.
Rule 1: Ifthe color of the fruit is "RED" and the size of the fruit is
"SMALL" then the fruit is cherry.
Rule 2: If the color of the fruit is "RED" and the size of the fruit is
"BIG" then the fruit is apple.
Rule 3: If the color of the fruit is
"GREEN" and the size of the fruit is "SMALL" then the fruit is
grape.
Rule 4: If the color of the fruit is
"GREEN" and the size of the fruit is "BIG" then the fruit is watermelon.
A common use
case of
It
supervised learning is to utilize historical data to predict statistically likely future
cvents. may use historical stock market information to anticipate upcoming Huctuations, or be
to hlter out
spam emails. employed
All
supervised learning techniques are a form of classifncation or
regression.
Classification techniques predict discrete responses-for example, whether
Or
an email isgenuine or spam,
whether a tumor small, medium, or large. Classification models are trained
is
to
classify data into
Categories. Applications include medical imaging, speech recognition, and credit scoring.
*gression techniques predict continuous responses-for example, changes temperature or Huctua-
in
Tions in electricity demand. Applications include forecasting stock prices,
and acoustic
signal handwriting recognition,
processing
W are
working on classification problem, we begin by determining whether the
a

binary classification problem, single training test item (instance) canproblembe divided
is binary
i . a In
a or

Classes-for example, if we want to determine whether an email is genuine or spam.only In a multiclass

Cation problem, it can be divided into more than two classes-for
classify example, if we
want to train a
roblem
an
image dog, as
cat, or other animal. Bear in mind that
a
muliclass classifhcation
a
Eenerally more challenging because it requires more
complex model.
a
 CHAPTER 1/INTRODUCTION TO MACHINE LEARNINC
10

Example:
ResearchersCreating
Algorithms that Can Analyze Works of Art
at the Art and Artificial Intelligence Laboratory at Rutgers University wanted to see whether a
as a human. They began by
computer algorithm could classify paintings by style, genre and artist as casily
identifying visual features for classifying a painting's style. The algorithms they developed classified the styles
of paintings in the database with 60% accuracy, outperfornming typical non-expert humans. The researchers
hypothesized that visual features are useful for style classification (a supervised learning problem). They used
classification algorithms trained on Google images to identify specific objects. They tested the algorithms on
more than 1,700 paintings from 66 different artists working over a span of 550 years. The algorithm readily
identified connected works, including the influence of Diego Velazquez's "Portrait of Pope Innocent X" on
Francis Bacon's "Study After Velazquezs Portrait of Pope Innocent X."

1.4.2 Unsupervised Learning

In unsuperviscd learning, the data is unlabeled, so the learning algorithm is left to find commonalities among
its input data. As unlabeled data is more abundant than labeled data, machine learning methods that facilitate
unsupervised learning are particularly valuable. The goal of unsupervised learning may be as straightforwar
as discovering hidden patterns within a dataset, but it may also have a goal of feature learning, which allow-
the computational machine to automatically discover the representations that are needed to classify raw data
dataset of custom
Unsupervised learning is commonly used for transactional data. You may have a large
ers and their purchases, but as a human you will likely not be able to make sense of what similar attribute-
can be drawn from customer profiles and their types of purchases. With this data fed into an unsupervise-
unscented soaps an
learning algorithm, it may be determined that women of a certain age range who buy
likely to be pregnant, and therefore a marketing campaign related pregnancy
to and baby products can E
targeted to this audience in order to increase their number of purchases.
Without being told a "correct answer, unsupervised learning methods can look at complex data that
more expansive and seemingly unrelated in order to organize it in potentially meaningful ways. Unsupervise
is often used for detection including fraudulent credit card purchases and recommend-
learning anomaly
systems that suggest what products to buy next. As an example, in unsupervised learning, untagged photos-
dogs can be used as input data for the algorithm to find likenesses and classify dog photos together (Fig. 1.

aatt

Figure 1.9 Dogs getting tagged among ditferent animals.

 1.4 TYPES OF MACHINE LEARNING
11

Most unsupervised learning techniques are a form of cluster analysis. In such analysis, data is partitioned
into groups based on some measure of similarity or shared characteristic. Clusters are
formed so that objects
in the same cluster are very similar and objects in difterent clusters are
very distinct.
Clustering algorithms
fall into two broad groups: (a) hard custering, where cach data point
belongs only cluster and (b) can
to one
clustering, where cach data point can belong to more than one cluster. Hard or soft clustering techniques soft
be used if you already know the possible data groupings.

Example 1.4: Hard Clustering

Using k-means clustering to site cell phone towers, a cell phone company wants to know the number and
placement of cell phone towers that will provide the most reliable service. For optimal signal reception,
the towers must be located within clusters of people. The workflow
number of clusters that will be needed. To evaluate this
begins with an initial guess of the
guess, the engineers compare service with three
towers and four towers to see how well they are able to cluster for each
scenario (in other words, how
well the towers provide service). A phone can only talk to one tower at a time, so this is a hard clustering
problem. The k-means
team uses
clustering because it each observation in the data as an
treats
having a location in space. It hnds a partition in which objects within each cluster are as close toobjecteach
O
other as possible and as far from objects in other clusters as
possible. This is illustrated in Fig. 1.10; the
small dots denote the customer's geometric location information and
1Se
suggested tower location.
bigger black dots represent the

hati
VIS
end
tos
15

Figure 1.10 Cell tower sites identified based on customer location.

ter running the algorithm, the team can accurately determine the results of partitioning the data into
three and four
clusters.
 CHAPTER 1/INTRODUCTION TO MACHINE LEARNIN
12

Example 1.5: Soft Clustering understand the genes involved

is analyzinggene expression data from microarrays to better
A team of biologists if it is actively involved in a cellular
division. (A gene is said to be "expressed"
in normal and abnormal cell data from two tissue samples
production.) The microarray contains expression
function such as protein certain patterns of gene expression are
to determine whether
want to compare the samples
The researchers cluster the data. Because
After preprocessing the data to remove noise, they
implicated in cancer proliferation. no single gene is likely
to belong to one cluster
be involved in several biological processes,
the genes can
visualize the clusters to identify
to the data. They then
s a m e
algorithm
only. The researchers apply fuzzy
c-means
a and circles
shows two lusters represented by hexagons
that behave in a similar way. Fig. 1.11 cluster.
groups of genes which gives the degree of belongingness
to each
Each element is associated with a membership
(0.97, 0.03)

(0.03, 0.97) OO
(0.47, 0.53)

and circles.
1.11 Two clusters represented by hexagons
Figure

concerned wi
1.4.3 Reinforcement Learning behaviorist psychology,
is an area of machine learning inspired by notion of cumulau
Reinforcement learning e n v i r o n m e n t so as to
maximize some

action in an
such as game theory,
ca
how software agents ought to take in other disciplines,
to its generality,
is studied many
multi-agent syster
reward. The problem, due simulation-based optimization,
information theory, literature, t
trol theory, operations
research, research and control
In the operations
statistics and genetic algorithms. programming
Swarm intelligence,
methods are studied is
called approximate dynamic
held where reinforcement learning

hree
Example 1.6 1.12). Two players
take play1ng on a
turns

the familiar child's game of tic-tac-toe (Fig. wins by placing three marks
Consider the other O's. A player
board. One player plays X's and three in a row, thc
by-three board fills with neither player getting aga
If the up
that w e a r e playing
tally, vertically, or diagonally. to lose, let us assume cor
nside

is a draw. Because a skilled player can play

never
In fact, let
so as us

incorrect and allows us to win. nperte

an imperfect player, one whose play

is sometimes the
uup
c o n s t r u c t a player
that will find
and losses to be equally bad for us.
How might we
draws
tions in its opponent's play and learn to maximize his chances of winning
1.4 TYPES OF MACHINE LEARNING 13

xOO
OXX
X
Figure 1.12 The tic-tac-toe game.

Although this is a simple problem, it cannot readily be solved in a satisfactory way through classical techniques.
For example, the classical minimax solution from game theory is not applicable here because it assumes a partic-
ular way of playing by the opponent. A minimax player would never reach a game state from which it could lose,
even if in fact it always wins from that state because of incorrect play by the opponent. Classical optimization
methods for sequential decision problems, such as dynamic programming, can compute an optimal solution for
any opponent, but require as input a complete specification of that opponent, including the probabilities with
which the opponent makes each move in each board state. Let us assume that this information is not available
for this problem, as is the case for the vast majority of problems of practical interest. Such information can be
estimated from experience, in this case by playing many games against the opponent. About the best one can do
with this problem is to learn a model of the opponent's behavior up to some level of confhdence, and then apply
dynamic programming to compute an optimal solution given the approximate opponent model (Fig. 1.13).

#
x's move

o's move

A MA x's move

AMMMM MAAAMA o's move

x's move

Figure 1.13 Tree-like structure derived from playing tic-tac-toe.

14 CHAPTER 1/INTRODUCTION TO MACHINE LEARNING

Case Study 1: Diagnosing Crop Disease with Machine Learning

An agricultural biotech client needed a way to diagnose crop discasesto help farmer community. Losses du
to crop disease, whether caused by fungi, bacteria, or viruses, can be disastrous to the economy and dramati
cally threaten the global population's access to nutrition.
Crop discase devastate natural ecosystems, leading to environmental problems such as habitat loss
can

for certain species. Additionally, in extreme cases, diseased crops can produce toxins that result in seriou
health problems for consumers.
The following are reasons why this crop disease problem was chosen as a case study. Accurate diagnosi
was the best way to target the exact solutions necessary to maintain crop health. This diagnosis cannot be
done without deep subject matter expertise. Moreover, accurately describing a crop's symptoms to a plant

over the phone is nearly impossible. The

solution to this problem was the development of a
pathologist
mobile application that allows farmers to photograph diseased crops and receive accurate diagnostics in
real-time. The diagnostic process occurs through the use of advanced image analytics powered by machine

learning
Advanced image analytics: The mobile application uses Google's Cloud Vision API to analyze each crop's
color, size, texture, and decay patterns, then references these data points against a library of 50,000 images.

These images are classified into categories and labeled. the mobile application is
Machine learning: Through pattern-recognition machine-learning techniques,
trained to diagnose crop disease in near real-time.
model management, along with DevOps support
Data engineering: A reusable platform enables data and
data cleansing, learning, and service
for multi-cloud infrastructure to assist data scientists in performing
at scale.
deployment of solutions, utilizing artificial intelligence analytics techniques

Case Study 2: Learning to Decode the Immune System

to Diagnose Disease
itself detect
The human immune system is an astonishing diagnostic system, continuously
adapting to

of human
disease in the body. Machine learning can enhance the scientifhc understanding
any
health and of
signal provide a foundation for a new generation of precise medical diagnostic and treatment options:
firm, along with Microsoft works towards
Adaptive Biotechnologies, a Seattle based bio-technological
and machine learning with recent breakthroughs
coupling the latest advances in artificial intelligence
and decoding the human immune systct
biotechnology to build a practical technology for mapping
T h e goal o f this research is to "create a universal blood test that reads a person's i m m u n e system to u c

earliest stag
of diseases including infections,and autoimmune disorders in their
cancers,
a wide variety of such a univets
on human health
when they can be most effectively diagnosed and treated". The impact
to disease would be, in a word, transformational. ru
blood test that reads a persons exposure and response has
the immune system can reveal what diseases the body currently
is fighting or crem
Simply, sequencing what the immune sys
contains the key information needed to read
fought. A blood sample, therefore,
is currently detecting.
MULTIPLE-CHOICE QUESTIONS
15

Summary
Machine learning is an area in
computer science
which involves teaching computers to do Supervised, unsupervised and reintorce-
things ment
learning are
types of machine
learning
naturally by learning through experience. algorithms.
According to Tom M Mitchell, A computer Supervised learning is also called learning with
program is said to learn from experience (E) with
a tutor, builds a model with known
respect to some class of tasks (1) and performance input and
measure (P), if its performance at tasks in 7, as corresponding outputs.
measured by ? improves with experience E. Unsupervised learning algorithms try to identify
patternsfrom the input of data samples.
Machine learning lies at the intersection of com- Reinforcement learning is an area of machine
puter science, engineering, and statistics and
often appears in other disciplines.
learning inspired by behaviorist psychology,
concerned with how software
Recommendation systems, recognition systems, agents ought to
take actions in an environment so as to maxi-
classifiers and prediction systems are some of the mize some notion of cumulative
reward.
applications that use machine learning.

Multiple-Choice Questions
1. Suppose your email program watches your 4. A sofrware
way of marking the emails as spam or not. designed to evaluate whether a
customer account is hacked or not is an
and based on these observations learns how to Cxample of
better filter spam. The task Tin this case is
(a) Unsupervised learning
(a) Classifying emails as spam or not spam
(b) Supervised classification
(6) Watching you label emails as spam or not
(c) Supervised regression
spam
(d) None of the above
The number of emails correctly classified as
spam/or not spam 5. Which of the following
examples would be
(d) None of the above classified as an example of unsupervised learn-
2. A toy
ing algorithm?
shop wants to predict the number of i. Given email labeled
Barbie dolls which would be sold the next
as
spam/not spam,
learn a spam filter.
three months. What
type of problem is this? ii. Given a set of news articles found on the
(a) Unsupervised web. group them in articles under finance,
(b) Supervised movies, and political domain.
(c) Reinforcement ii. Given a database of customer data, automat-
(d) None of the above ically discover market segments and group
3. customers into different market
What kind of learning is segments.
required to group a set iv. Given a database of patients
of unlabeled dolls
as Barbie, Frozen, and Ochers? diagnosed as
(a) Unsupervised learning either having diabetes or not, learn to clas-
sily new patients as having diabetes or not.
(b) Supervised learning
(a) Both i and iv
(c) Reinforcement learning
(b) Both i and i
(d) None of the above
(c) Both ii and ii
(d) Both ii and iv
16 CHAPTER 1//NTRODUCTION TO MACHINE LEARNING

Very Short Answer Questions

1. Define machine learning as given by Tom 4. How is hard clustering different from soft
Mitchell. clustering?
5. Define reinforcement learning with an
2. What is supervised learning?
3. What is unsupervised learning?
example.

Short Answer Questions

1. Explain machine learning with respect to (a) Establishing an appropriate algorithm that
would identify research groups working in
designing a chess game by identifying the tasks
(1), experience (E), and pertormance measure various domains.
(P). (b) Computing the returns on investment
2. Illustrate with a simple example how super- when the initial values of investments are EA
vised learning can be used in handling loan given.
defaulters. (c) Identifying whether a breast lump is malig-
3. Organize the following activities under super- nant or benign based on standard data

vised or unsupervised learning. State appropri sample taken from University of California
ate reasons for the same. Irvine reposito
4. Match the technique to the application.

Credit card fraud detection Supervised classification

Word frequency of afeatured article Supervised regression

Identifying whetheramailis spam or ham Unsupervised learning
Predicting the price ofa stock Outlier analysis

Review Questions

1. Define machine learning. 5. After a model is built using supervised learning

what do you do with the learned model?
2. Name some applications that require machine
learning techniques. 6. What is cluster analysis?
3. What are the different types of machine 7. How is learning accomplished in reinforcemen
learning algorithms? learning
4. How is supervised learning different from
unsupervised learning techniques?
Answers

Multiple-Choice Questions
1. (a) 2.(b) 3. (a) 4.(b) 5. (c)