Data mining

Francia Theresa L. Oliver

 MACHINE LEARNING
Introduction to Machine Learning
➥used by search engines such as Google and Bing to
rank web pages or to decide which advertisement to
show to which user.
➥used by social networks such as Facebook and
Instagram to generate a custom feed for every user to
tag the user by the picture that was uploaded.
➥banks to detect whether an online transaction is
genuine or fraudulent and e-commerce websites such
as Amazon and Flipkart to recommend products that
we are most likely to buy.
➥Gmail, Yahoo, and Hotmail use machine learning to
decide which emails are spam and which are not.
Machine Learning
➥Branch of Artificial Intelligence
➥But it is very difficult to make rules for programs
such as photo tagging, classifying emails as spam or
not spam, and web page ranking. The only solution to
accomplish these tasks was to write a program that
could generate its own rules by examining some
examples (also called training data). This approach
was named Machine Learning.
Machine Learning
AI researchers believe that
this goal can be achieved
through machine learning
algorithms that try to mimic
the learning processes of the
human brain.
build an Artificial
Intelligence (AI) platform
that is as intelligent as
the human mind.
Applications of Machine Learning
1. Virtual Personal Assistants
➥such as Siri, Alexa, or Google Assistant
⇢these assistants can be trained for certain tasks like
‘Set an alarm for 5 AM next morning’, ‘Remind me to
visit doctor tomorrow at 6 PM’, and so on.
➥Smart Speakers like Amazon Echo and Google
Home are the outcomes of this innovation.
➥Above-mentioned assistants use machine
learning to achieve these Objectives.
2. Traffic Predictions
➥such as Google maps
⇢it uses machine learning to predict the expected time
of arrival at the destination and also to model traffic
congestion on real time basis.
WEEK 1-2
3. Online Transportation Networks
➥book cabs by using mobile apps like Grab & Uber.
⇢these apps estimate the price of the ride by using
machine learning.
⇢they also use ML to determine price surge hours by
predicting the rider’s demand
4. Video Surveillance
➥such as CCTV cameras
⇢ML based video surveillance systems can even detect
crime before it happens.
⇢It can track unusual behavior in people such as
standing motionless for an unnaturally long time,
hesitancy, or napping on benches and such-like.
⇢the system can thus alert human attendants, for
suitable response.
5. Social Media Services
➥such as Facebook
⇢personalizing news feed in order to better
advertisement targeting over social media
⇢to show news feed to the user based on his or her
interests by considering items clicked earlier by
that user
⇢a list of Facebook users are suggested for you to
become friends with.
⇢The Face Recognition feature of Facebook also uses
ML to tag the friends in a picture. Facebook checks
the poses and projections in the picture, notices the
unique features, and then matches them with the
people in your friends list.
➥Pinterest also uses ML for computer vision to
identify the objects (or pins) in the images and
recommend similar pins accordingly.
6. Email Span and Malware Filtering
➥have inbuilt machine learning to identify spam
emails.
➥ the system learns and identifies new mail as
spam or not, automatically.
7. Online Customer Support
➥Many sites these days offer the surfer the option
to talk with them.
➥In any case, very few sites have a live official to
answer your questions. In most cases, you
converse with a chatbot.
➥These bots extract data from the site and
present it to clients through machine learning.
8. Search Engine Result Refining
➥Google and other search engines use machine
learning to improve search results for you.
9. Product Recommendations
➥Whenever we make an online purchase on
websites such as Amazon or Flipkart or
similar, we usually keep receiving emails from
them with shopping suggestions. They also
recommend us items that somehow match our
tastes.
10. Online Fraud Detection Classification of Machine
➥Machine learning is also helping in making Learning Algorithms
cyberspace more secure and tracking
monetary frauds online. Machine Learning
➥PayPal is using ML for protection against money
laundering. The company uses a set of ML
tools that helps them to compare millions of
transactions taking place and distinguish
between legitimate or illegitimate transactions Supervised Reinforcement
Unsupervised
taking place between the buyers and sellers. (Regression/
Classification)
(Clustering)
(Algorithm learns to
react to environment)

11. Medicine
➥With the advent of automation, medical records
are now available in electronic form.
➥The ML algorithms are helping doctors to
understand diseases in a better manner by
turning the medical records into medical
knowledge.
12. Handwriting Recognition
➥ML can not only recognize handwriting but also
read different ones. So, it is a versatile tool for
many applications.
➥For instance it can be used to route postal mail
all over the country once it has been trained
to read addresses written in anyone’s
handwriting.
Site: https://quickdraw.withgoogle.com/#
13. Machine Translation
➥We use Google Translate that translates
text/website instantly between 100 different
human languages as if by magic.
14. Driverless cars and autonomous helicopters
➥One such application is self-driving cars or
autonomous helicopters.
➥It takes years of experience for a person to
become a good driver and much of this is
intuitive. The only possible solution is
machine learning, i.e., having a computer
program that can learn by itself how to drive
a car or fly a helicopter.
Supervised Learning
➥Supervised learning can be used for two types of
problems: Classification and Regression.
➥machine learning method in which models are
trained using labeled data.
➥models need to find the mapping function to
map the input variable (X) with the output
variable (Y). Y=f(X)
➥needs supervision to train the model, which is
similar to as a student learns things in the
presence of a teacher.
📌 The example discussed below is an example of
supervised machine learning and the term ‘supervised’
signifies the fact that the dataset with the ‘right
answers’ is given to the algorithm. The given example
is also a case of regression problem. In a regression
problem, the system predicts a continuous– valued
output (here, it is the price of the house).

Defining Machine Learning

➥Arthur Samuel (1959) coined the term machine
learning and defined it as: ‘the field of study that gives
computers the ability to learn without being explicitly
programmed.’ This is an informal and old definition of
machine learning.
➥In 1998, Tom Mitchell redefined the concept of
machine learning as ‘[A] computer program is said to
learn from experience E with respect to some class of
tasks T and performance measures P, if its
performance at tasks in T, as measured by P, improves
with experience E.’ For instance, let’s say there is an
email program which tracks the emails that a person
marks as spam or not spam and based on that learns
how spam can be filtered in a better way. Thus,
Classification of emails as spam or not spam is the
Task T. Tracking the user and marking emails as spam
or not spam becomes Experience E. The number of
emails correctly classified as spam or not spam is
Performance P.
📌 The example below is an example of Classification. For example, the age of the patient can also be
The term ‘classification’ signifies that the system has considered instead of just knowing the size of tumor
only. In that case, our dataset would look like as shown
to predict the output as a discrete value, i.e., one or
in Figure 2.3.
zero (either benign or malignant in the above
example). It should be noted that in a classification
problem, the output can have more than two possible
values. For example, there may be three types of
breast cancers and one can try to predict the discrete
value, i.e., zero, one, or two.
Here, zero may represent a benign tumor or not
harmful cancer, one may represent type one cancer
and the discrete value two may indicate a type two
cancer.
Hence, in a classification problem, we may have N
classes in the output where N is always a finite
Let’s suppose a person has a tumor, of size and age
number. as depicted by B in Figure 2.4. In this dataset, the ML
From the given example, only one attribute or algorithm is able to fit a straight line to separate out
feature namely the tumor size has been used with the the benign tumors from the malignant tumors as
aim to predict whether the type of tumor is benign or shown in Figure 2.5.
malignant. In other machine-learning situations, there Thus, according to the o ML algorithm, a straight
line can, hopefully, help us determine a person’s tumor
can be more than one attribute or feature. For
by separating out the tumors. And if a person’s tumor
example, the age of the patient can also be considered falls in this benign area, then the type of cancer is
instead of just knowing the size of the tumor only. more likely to be benign than malignant.
' In this example, two features such as the patient’s
age and tumor size have been considered. However,
we can increase the number of features to further
increase the accuracy of prediction. It is important to
note that as more and more relevant features are
added to a model, the model will become more
complex but the accuracy may increase. Commonly,
machine learning algorithms have more than one
feature.

Problem 1:
For each customer account, you’d like to have a
software to check the account and decide if it has been
compromised (hacked).

Problem 2:
You have a large stock of similar goods. You wish to
know that in the next three months how many of these
goods will sell.
 Unsupervised Learning 📌Other Examples Of Unsupervised Learning Applications
➥Unsupervised learning, also known as unsupervised ➥Organize Computing Clusters
machine learning, uses machine learning algorithms ⇢The geographic location of servers is decided
to analyze and cluster unlabeled datasets. These on the basis of clustering of web requests
algorithms discover hidden patterns or data groupings received from a particular region of the
without the need for human intervention. world. The local server will then contain only
➥Its ability to discover similarities and differences in the data frequently accessed by people of
information makes it the ideal solution for exploratory that region.
data analysis, cross-selling strategies, customer ➥Social Network Analysis
segmentation, and image recognition. ⇢Social network analysis is conducted to
➥allows us to approach problems with little or no idea create clusters of friends based on the
about what the results will look like. We can derive frequency of communication among them.
structure from data where we don’t necessarily know Such analysis reveals the links among the
the effect of the variables. users of any social networking website.
➥data is not labeled, which means that there is no ➥Market segmentation
output attribute. We only have input attributes and on ⇢Sales organizations can cluster or group
the basis of the values of input attributes grouping or their customers into different segments on
clustering is performed on the input data to group the basis of their previous billed items. For
them into similar classes. example, a big superstore may want to send
➥We can only construct the clusters of data based on an SMS about grocery items specifically to
relationships among the variables in the data. With its customers of grocery instead of sending
unsupervised learning, there is no feedback about the that SMS to all its customers. Not only is it
results predicted, i.e., there is no teacher to correct cheaper but also better; after all it might be
you. an irrelevant irritant to those who only
purchase clothing from the store. The
📌 Examples of Unsupervised Learning: grouping of customers into different
segments based on their purchase history
➥Google news as depicted in Figure 3.1 is an
will help the store to target the correct
excellent example of clustering that uses
customers for increasing sales and
unsupervised learning to group news items
improving its profits.
based on their contents.
➥Astronomical data analysis
Google has a collection of millions of news
items written on different topics and their ⇢Astronomers use large telescopes to study
clustering algorithm automatically groups galaxies and stars. The patterns in light or
these news items into a small number that grouping of lights received from different
are somehow similar or related to each parts of the sky help to identify different
other by using different attributes, such as galaxies, planets, and satellites.
word frequency, sentence length, page
count, and so on Reinforcement Learning
➥Astronomical data analysis used in applications like
computer games (where the machine plays with a
human), driverless cars, robot navigation, etc.
➥It works through trial and error, and the machine
selects those actions that yield the greatest rewards.

Machine Learning

AGENT ENVIRONMENT ACTIONS

It is used for It defines the It defines the
learning and outer world with tasks to be
decision making. which the agent performed by the
The agent chooses interacts. agent.
actions that
maximize some
specified reward
metric over a given ➥reinforcement learning allows machines
amount of time
and software agents to automatically
determine the ideal behavior within a
specific context, in order to maximize its
performance. Simple reward feedback is
required for the agent to learn its behavior
and this is known as the reinforcement
signal.