Chapter 8

Learning from Examples

 What is Learning ?
• A distinct feature of intelligent agents is their ability to LEARN from experience.
• A learning agent is able to produce new knowledge using his experience and
internal knowledge.
• Any component of an agent can be improved by learning from data.

© Bushra Alhijawi 2
 Learning Elements
• Machine learning research has produced a large variety of learning elements.
• Major issues in the design of learning elements:
• Which components of the performance element are to be improved?
• What representation is used for those components?
• What kind of feedback is available:
• supervised learning.
• reinforcement learning.
• unsupervised learning.
• semi-supervised learning.
• What prior knowledge is available?

© Bushra Alhijawi 3
 Machine Learning Types
• Supervised learning → uses labeled datasets
to train algorithms to classify data or predict
outcomes accurately.
• Unsupervised learning → uses unlabeled
datasets to train algorithms to analyze and
discover hidden patterns or data groupings
without the need for human intervention.
• Semi-Supervised learning → similar to
supervised learning but uses labeled and
unlabeled datasets to train algorithms.
• Reinforcement learning → a behavioral ML
model that learns as it goes by using trial and
error
© Bushra Alhijawi 4
 Supervised Learning
• The dataset is the collection of labeled examples
involving features and labels.

• Supervised learning methods can be applied to:

• Regression problems → Prediction of a continuous
value.
• Ex. Sales prediction, trend prediction.
• Classification problems → Prediction of discrete
classes (categories) to which samples belong.
• Ex. Spam email, movie to categories.

© Bushra Alhijawi 5
 Supervised Regression
Features Supervised Label
ML Algorithm
Feature 1 … Feature n Goal 1

Feature 1 … Feature n Goal 2

Feature 1 … Feature n Goal 3

Name Year Transmission K_Driven Engine

price
Hyundai
2015 Manual 41000 1582 CC
Creta Supervised regression algorithms can 12500
Honda predict and infer causal relationships
2011 Manual 46000 1199 CC 4500
Jazz V between independent and
Audi A4 2013 Automatic 40670 1968 CC dependent variables.
17740
© Bushra Alhijawi 6
 Supervised Regression - Linear Regression
• Linear regression learns a
model which is a linear
combination of features of the
input example.
• Regression searches for
relationships among variables.

The core idea is to obtain a line

that best fits the data.

𝑓 𝑥 = 𝑤𝑥 + 𝑏
© Bushra Alhijawi 7
 Supervised Regression -Polynomial Regression

If we use a simple linear

regression in this data, the
linear regression line will not fit
very well. It is challenging to fit
a linear regression line in the
above graph with a low error
value.

© Bushra Alhijawi 8
 Supervised Regression -Polynomial Regression

In such a case, the polynomial

regression is used to fit a
polynomial line to achieve a
minimum error or minimum cost
function.

© Bushra Alhijawi 9
 Supervised Regression -Polynomial Regression
• Polynomial regression fits a nonlinear relationship between
dependent and independent variables.

© Bushra Alhijawi 10
 Supervised Classification
Features Supervised Label
ML Algorithm
Feature 1 … Feature n Goal 1

Feature 1 … Feature n Goal 2

Feature 1 … Feature n Goal 3

Name Year Transmission K_Driven Engine

Hyundai In classification, abstract classes are Price category

2015 Manual 41000 1582 CC
Creta formed to delimit and order data
Honda meaningfully. For this purpose, Mid
2011 Manual 46000 1199 CC
Jazz V objects are obtained based on
Audi A4 2013 Automatic 40670 1968 CC similar characteristics and structure. Very Low

© Bushra Alhijawi High 11

 Supervised Classification - Logistic Regression
• Logistic Regression is used when the dependent
variable(target) is categorical.
• Logistic Regression (also called Logit Regression) is
commonly used to estimate the probability that an
instance belongs to a particular class.
• Ex. what is the probability that this email is spam?
• If the estimated probability is greater than 50%, the
model predicts that the instance belongs to the positive
class (labeled “1”), else it predicts that it belongs to the
negative class (labeled “0”). This makes it a binary
classifier.
• The logistic is a sigmoid function (i.e., S-shaped) that
outputs a number between 0 and 1.

© Bushra Alhijawi 12
 Supervised Classification - k-Nearest Neighbors
• The KNN algorithm assumes that
SIMILAR things exist in close
proximity.
• As the value of K increases, the
predictions become more stable
due to majority voting /
averaging, and thus, more likely
to make more accurate
predictions (up to a certain
point).

© Bushra Alhijawi 13
 Supervised Classification – Decision Tree
• Decision tree is an acyclic
graph that can be used to
make decisions.
• A non-parametric
supervised learning
method.
• Very powerful algorithms,
capable of fitting complex
datasets.
• The goal is to create a model
that predicts the value of a
target variable by learning
simple decision rules inferred
from the data features.
Image Source
© Bushra Alhijawi 14
 Supervised Classification – Decision Tree

Example Source
© Bushra Alhijawi 15
 Performance Evaluation – Supervised Learning
• Evaluating a machine learning model is important to assess its performance,
identify potential problems, and make improvements.
• A well-evaluated model can help improve its accuracy, generalizability, and
reliability.
• Performance evaluation steps:
• Data Sampling.
• Define the evaluation metrics.
• Train the machine learning algorithm.
• Test the machine learning model.
• Analyze the results to identify potential problems.
• Refine the model.

© Bushra Alhijawi 16
 Evaluation – Data Sampling
• Divide the data into training, validation, and testing sets.
• The training set is used to train the model.
• The validation set is used to optimize the hyperparameters and prevent overfitting.
• The testing set is used to evaluate the final performance of the model.
• Data sampling techniques:
• Hold-Out → The original dataset is randomly divided into two subsets: the training
set and the testing set.
• K-fold Cross-Validation → The original dataset is divided into K subsets. The model is
then trained and tested k times, with each fold serving as the testing set once and
the remaining folds serving as the training set.
• Leave-One-Out → A type of cross-validation method where the original data is
divided into n-folds, where n is the number of samples in the dataset.
• Boot Strapping (Bagging) →Creating multiple new datasets by randomly sampling
from the original dataset with replacement and then estimating the performance or
estimate on each of these new datasets.
© Bushra Alhijawi 17
 Evaluation – Data Sampling
• Hold-Out method

• K-fold Cross-Validation method

Source

© Bushra Alhijawi 18
 Evaluation – Data Sampling
• Leave-One-Out method

• Boot Strapping

© Bushra Alhijawi 19
 Evaluation Metrics – Regression
• Mean squared error (MSE): the average of squared differences between the
predicted output and the true output.

• Mean Absolute Error(MAE): the average of the absolute differences between the
predicted output and the true output.

© Bushra Alhijawi 20
 Evaluation Metrics – Classification
• Confusion matrix → A summary of the number of correct and incorrect
predictions made by the model, comparing the predicted labels to the true labels
in the data.
• True Positives (TP): Positive records are
correctly classified as positive.
• True Negatives (TN): Negative records are
correctly classified as negative.
• False Positives (FP): Negative records are
misclassified as positive.
• False Negatives (FN): Positive records are
misclassified as negative.

© Bushra Alhijawi 21
 Evaluation Metrics – Classification

© Bushra Alhijawi 22
 Evaluation Metrics – Classification

© Bushra Alhijawi 23
 Evaluation Metrics – Classification

© Bushra Alhijawi 24
 Evaluation Metrics – Classification
• How to build the confusion matrix?

© Bushra Alhijawi 25
 Unsupervised Learning
• In unsupervised learning, the dataset is a collection of unlabeled
examples (ONLY Features).
• Unsupervised learning methods determined hidden patterns in
data without initial patterns and relationships being known.

© Bushra Alhijawi 26
 Unsupervised Learning
Features Unsupervised
ML Algorithm
Feature 1 … Feature n

Feature 1 … Feature n
Internal
Similarity

Feature 1 … Feature n

Name Year Transmission K_Driven Engine

Hyundai Category
2015 Manual 41000 1582 CC
Creta
Honda Cluster 1
2011 Manual 46000 1199 CC
Jazz V
Cluster 2
Audi A4 2013 Automatic 40670 1968 CC

© Bushra Alhijawi 27
 Semi-Supervised Learning
• In semi-supervised learning, the dataset contains both labeled and unlabeled
examples.
• Usually, the quantity of unlabeled examples is much higher than the number of
labeled examples.
• The ML algorithm is trained using the labeled data and predicts the unlabeled
data to attach the label with every data sample (Pseudo-labeled data). Then, a
new model can be trained with a mixture of labeled and pseudo-labeled data.

© Bushra Alhijawi 28
 Reinforcement Learning
• In Reinforcement learning, the machine “lives” in an environment and can
perceive the state of that environment as a vector of features. The machine can
execute actions in every state. Different actions bring different rewards and could
also move the machine to another state of the environment.
• Reinforcement learning → always looks for the BEST BEHAVIORS.
• The agent selects the BEST action from all options presented in the environmental
state, and based on that selection, receives reward/risk.

© Bushra Alhijawi 29
 Machine Learning Applications
• Online Customer Support
Several websites nowadays offer the option to chat with customer support
representatives while navigating the site. In most cases, you talk to a chatbot. These
bots tend to extract information from the website and present it to the customers.
They tend to understand the user queries better and serve them with better
answers.
• Search Engine Result Refining
Search engines use machine learning to improve search results. Every time you
execute a search, the algorithms watch how you respond to the results.
• Product Recommendations
The product recommendations are based on your behavior with the website/app,
past purchases, items liked or added to the cart, brand preferences, etc.

© Bushra Alhijawi 30
 Machine Learning Applications
• Virtual Personal Assistants
Ex., Siri, Alexa, and Google. Machine learning collects and refines the
information based on your previous involvement with them.

• Traffic Predictions
Machine learning in such scenarios helps estimate the regions where
congestion can be found based on daily experiences.

• Social Media Services

People You May Know, Face Recognition, Similar Pins, and more.

• Email Spam and Malware Filtering

© Bushra Alhijawi 31
Any Question
www.psut.edu.jo

www.psut.edu.jo
Call: (+962) 6-5359 949
Fax: (+962) 6-5347 295
[email protected]

Email: [email protected]
Princess Sumaya University for Technology
Amman 11941 Jordan
P.o.Box 1438 Al-Jubaiha