MACHINE

LEARNING
LEARNING OUTCOMES

• Define machine learning.

• Differentiate between data mining and machine learning.
• Identify the benefits of machine learning.
• Explain the limitations of machine learning.
• Identify the skills required of a machine learning engineer.
• Differentiate between supervised and unsupervised machine learning.
• Identify applications of machine learning.
• Discuss the ethical considerations in machine learning.
INTRODUCTION

• In learning unit 4, we discussed the steps in the data science process.

• And we mentioned that the modelling step makes use of machine learning
algorithms.
• In this learning unit, we will look closely at this field without getting into the actual
programming involved.
DATA MINING AND MACHINE LEARNING

• Data mining, also known as knowledge discovery in data (KDD), is the

process of uncovering patterns and other valuable information from large data
sets.
• Machine learning is a subset of [Artificial Intelligence] that enables machines to develop
problem-solving models by identifying patterns in data instead of leveraging explicit
programming.
• Data mining is the process of discovering patterns and extracting insights from
large datasets, while machine learning focuses on developing algorithms and
models that learn from data and make predictions or decisions.
DATA MINING AND MACHINE LEARNING
CONT…
• In data mining, the process is human-driven.
• A set process is followed to arrive at the answer, which is only as good
as the user that inputs the process parameters.
• But in machine learning, the applied algorithms can use data to learn
and become more effective.
• Machine learning is a subset of artificial intelligence (AI).
ARTIFICIAL INTELLIGENCE

• Artificial intelligence leverages computers and machines to mimic the

problem-solving and decision-making capabilities of the human mind.
MACHINE LEARNING ALGORITHMS ARE
GOOD AT:
• Classification of objects, for example, determining if an email should be
considered spam or what kind of animal is depicted in a photograph;
• Predicting future values, for example, how a city will consume much power in
winter;
• Anomaly detection, for example, detecting unusual network traffic.
BENEFITS OF MACHINE LEARNING

1. A more personalised experience for customers, based on their

preferences. An example of this is the Netflix recommendation system.
2. Business processes can be automated to make the business more
productive.
3. Predictions can be made of vital areas like product demand, which can
then be used to plan various business areas.
4. Because the machine learning algorithms can learn from changes in
data, using the algorithms will also allow the business to be more
adaptable to changes in the market.
BENEFITS OF MACHINE LEARNING CONT…

5. Information systems can be kept safe by using anomaly detection and

even face recognition.
6. For companies with production plants, machine learning can predict
when maintenance will be required.
LIMITATIONS OF MACHINE LEARNING

1. Machine learning algorithms compute statistical values and not absolute truths.
2. For machine learning to work well, we need lots of good data.
3. Machine learning results are often hard to fully explain.
4. It can be hard to diagnose and fix errors in complex machine learning systems if something
goes wrong.
5. Machine learning algorithms need to be trained on data first, so a system cannot immediately
make good predictions before it is trained.
6. A machine-learning algorithm can learn the wrong things due to noise in the original data set.
SUPERVISED AND UNSUPERVISED
MACHINE LEARNING
• There are two major classes of machine learning algorithms –
i. supervised and
ii. unsupervised.
SUPERVISED LEARNING

• In supervised learning, a training data set is provided to the algorithm.

• The training data has inputs and corresponding outputs.
• Based on the training data, the algorithm learns how to classify or predict.
• For example, if we are training a machine-learning algorithm to classify
images according to the type of animal that appears in each image, the
training data set would contain images with labels such as “dog” and “zebra”.
• The goal of supervised learning is then to correctly determine what the
outcome should be for new data.
UNSUPERVISED LEARNING
• In unsupervised learning, the algorithm still requires data to learn from, but we
don’t need to provide the corresponding outputs.
• Algorithms in this category can group data based on common features or uncover
relationships between different variables in the data.
• The goal of unsupervised learning is to determine what is interesting in the data
set that is presented to the algorithm.
APPLICATIONS OF MACHINE LEARNING

• 1. Self-driving vehicles are getting quite advanced and can for example, detect
when a human is crossing the road.
• 2. Companies use machine learning to recommend products and services to their
customers.
• 3. Virtual personal assistants can listen to voice inputs and respond accordingly.
• 4. Traffic can be predicted when you use a navigation app.
• 5. Automated chatbots can answer many customer support queries, improving
the response time to queries.
APPLICATIONS OF MACHINE LEARNING
CONT…
• 6. Machine learning algorithms can determine the emotion of a system user.
• 7. Text can be automatically translated using machine learning.
• 8. In the insurance industry, machine learning can be used for fraud detection,
making processing claims much faster.
• 9. Stock market prices can be predicted using machine learning algorithms.
• 10. In real estate, property valuations can be done using machine learning.
• 11. Plagiarism in student work can be detected by means of machine learning
algorithms.
SKILLS OF A MACHINE LEARNING ENGINEER
SKILLS OF A MACHINE LEARNING ENGINEER
CONT…
• Necessary mathematical skills include linear algebra and multivariate calculus.
• Statistical skills that are necessary include probability and various distributions.
• Basic programming skills are essential since most machine learning work is done in a
programming language like Python.
• Algorithms in machine learning, categorised as supervised or unsupervised algorithms.
• Neural networks are modelled after the human brain.
• Natural language processing allows machines to understand language.
• Lots of libraries exist that make building a machine learning algorithm much easier.
• Software engineering as well as system design.
ETHICAL CONSIDERATIONS

• Ethics is based on well-founded standards of right and wrong that

prescribe what humans ought to do, usually in terms of rights,
obligations, benefits to society, fairness, or specific virtues.
• It is all about what is the right thing or wrong thing to do.
• So, when we talk about ethics in machine learning, it is about doing the
right thing with our data and algorithms.
ETHICAL CONCERNS IN MACHINE LEARNING

i. Bias - A bias is a tendency, inclination, or prejudice toward or against

something or someone.
ii. Privacy - Machine learning algorithms use data that is often private to the
client/customer/user whom the data belongs to. So, privacy is an important
concern in machine learning.
iii. Deepfakes - Machine learning technology has made it possible to create deep
fakes. A deepfake is an image or video that was generated using a machine
learning algorithm to create footage that is not real.
PRINCIPLES FOR ETHICAL AI
PRINCIPLES FOR ETHICAL AI CONT…

• 3.2.1 “Beneficence: promoting well-being, preserving dignity, and sustaining the

planet”
• The first principle is that AI should be used to improve society and the world.
• 3.2.2 “Non-maleficence: privacy, security and ‘capability caution’”
• The second principle is around not harming. So, AI should not be used to violate
the privacy of a user, for example.
• 3.2.3 “Autonomy: the power to decide (to decide)”
• When implementing AI systems, this must not impede the ability of humans to
make their own choices.
PRINCIPLES FOR ETHICAL AI CONT…

• 3.2.4 “Justice: promoting prosperity, preserving solidarity, avoiding unfairness”

• It is important that introducing AI into the world must not introduce new ways of
discriminating against people. Instead, the decisions made by AI must be fair.
• 3.2.5 “Explicability: enabling the other principles through intelligibility and
accountability”
• When we discussed the limitations of machine learning, we already noted that one
cannot always explain how an algorithm reached a decision. Besides not being
popular, this also means we don’t know who to hold accountable for AI decisions. If
the algorithm could produce an understandable explanation, then we could know if
the creators introduced bias, for example.
A FINAL WORD

• We hope that you have enjoyed this exploration of the world of data.
• And that you have learned many skills that you will find helpful in your
career.
• In technology, things are ever-changing.
• But, this is only the start.