Scribd
Upload
47 views

CSC413-Machine Learning and Data Mining

This document provides an introduction to machine learning and data mining. It defines artificial intelligence as making computers able to do things that people are currently better at. It also discusses the need for investigating machine learning mechanisms and algorithms. Key concepts covered include supervised vs. unsupervised learning, characteristics of learning agents, applications of data mining, and some machine learning concepts like Occam's razor and overfitting.

Uploaded by

Kc Mama
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
47 views

CSC413-Machine Learning and Data Mining

This document provides an introduction to machine learning and data mining. It defines artificial intelligence as making computers able to do things that people are currently better at. It also discusses the need for investigating machine learning mechanisms and algorithms. Key concepts covered include supervised vs. unsupervised learning, characteristics of learning agents, applications of data mining, and some machine learning concepts like Occam's razor and overfitting.

Uploaded by

Kc Mama
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 7

1

CSC413: Artificial Intelligence (AI)


Machine Learning & A. A. Habila,
Department of
Computer Science-
Federal University Lafia
Monday, 20
February 2017

Data Mining
Monday, 20
February 2017 2

We can remember that AI can also be define as:

The study of how to make computers do things at


which, at the moment, people are better.

From the above, we can safely make a quick comparison of computers learning ability with
that of humans which will definitely reveal the inferiority of computers’ capacity as against
humans.
There is increasing need within the AI research community to intensify investigation into
learning mechanisms as well as develop learning algorithms as part of growing AI.
We are interested in understanding the theoretical underpinnings of learning as it happens in
nature and see how they form the basis for machine learning algorithms and their
applications.
We can think of questions like:
 What is learning?
 What learning theories are available and their possible applications to ML?
A. A. Habila, Department
 What is the difference between supervised and unsupervised learning? of Computer Science-
Federal University Lafia
3

A. A. Habila, Department Monday, 20


of Computer Science- February 2017
Federal University Lafia

The apple sorting classifier

 Features-size and colour

 Size is taken to be the diameter of the apple

 Colour: 0: green; 1:red

 The task of this agent is to generate a function from the collected classified data that maps
the features above to a discrete merchandise {A, B}

At this point, we can say that a Learning Agent (LA) is a function which maps a feature vector to a
discrete class value or in general a real number.

A function of this nature does comes into existence or experience changes during the learning phase
and is influence by the training data.

Machine Learning (ML) is the study of computer algorithms that improves automatically through
experience.

An agent is an LA if it improves its performance as can be measured by suitable criterion on new ,


and unknown data over time, after it has seen many training examples.
4

Characteristics of a Learning Agent


I. Task: Learn a mapping
II. Variable agent: the class of all learnable functions
III. Training Data: the knowledge which the LA needs to extract
and learn
IV. Test Data: this is use to evaluate whether the trained agent can
generalize well from the training data.
V. Performance Measure: this is an evaluation to determine the
quality of an agent. E.g. the percentage of correctly classified
apples.
A. A. Habila, Department
Monday, 20
of Computer Science-
February 2017
Federal University Lafia
5

A. A. Habila, Department Monday, 20


of Computer Science- February 2017
Federal University Lafia

There exist situations where we want the extracted knowledge to be presented in a readable manner for humans
as well. Data mining is about acquiring knowledge from data as well as its representation and applications.

Concisely, we can define Data Mining as the process of acquiring knowledge from data as well as its
representation and application.

Statistics and ML are very important methods and are expected to be applicable to very large amount of data at
a reasonable cost.

Areas of current applications

 Advertisement and Marketing

 Customer Relationship Management (CRM)

 Education and e-learning

 Internet and Search Engine etc.

Data Analysis

Statistics provides a number of ways to describe data with simple parameters.

We know of mean, standard deviation, statistical dependence (correlation) for multidimensional data,
covariance etc.
6

Monday, 20
February 2017

Some Machine Learning & Data Mining Concepts


Occam’s razor: for two scientific theories which explains the
same situation equally well, the simplest theory is preferred.
E.g. Algorithms that are compact, decision trees that are shorter
and are more easily understood by humans.
Overfitting effect: for a given learning algorithm, that is a
learning agent A, agent A is overfit to the training data if there is
another agent A’ whose error on the training data is larger than
that of A but whose error on the whole distribution of data is
smaller than the error of A.
Lazy Learning (Memory-based learning) Vs Eager Learning
A. A. Habila, Department
of Computer Science-
Federal University Lafia
7

A. A. Habila, Department Monday, 20


of Computer Science- February 2017
Federal University Lafia

End of Week-9!

You might also like