Lecture 1
What are we going to learn today?
- What is machine learning?
- What are supervised and unsupervised machine learning?
- Which are the most common types of machine learning problems?
- Which are the basic steps of the CRoss Industry Standard Process for data mining
(CRISP-DM)?
Machine learning is the field of study that gives computers the ability to learn without being
explicitly programmed
Machine learning
Assume that you are iterating over and over again an exercise
What should be constant in your exercise?
- Learning! - machine learning applies strategies and algorithms, combined with data
and statistics
- Improving! - machine learning applies statistical indices to measure the overlap
between ML prediction and expected result
When you are doing it, it is human learning
When a machine does it, it is machine learning!
An example of supervised learning
Supervised learning - classification
Given a labelled dataset, the model learns to
predict new examples
An example of unsupervised learning
Unsupervised learning - clustering,
dimensionality reduction, anomaly detection
and novelty detection
Given a dataset, without labels, the model
learns to use to cluster/group similar data
,CRISP-DM process model
Business understanding in the CRISP-DM process
Determine business objectives and success criteria
Business objectives and measures to evaluate the results have to be established
Business objectives:
● What is the customer’s primary objective?
● Increase the number of loyal customers
● Selling more of a certain product
● Have a positive marketing campaign
,Business success criteria:
● Objective measure to establish success (e.g. return of investment)
Main steps in a data mining project
1. Define the goals:
Business and data mining experts together have to define the goals. For each goal a
measure must be defined to understand its success
2. Obtain the models:
Pre-process the data, apply data mining algorithms
3. Evaluate results
Use the pre-specified measures to evaluate the models
4. Deploy:
If the evaluation is successful, the model can be deployed
Costs & benefits
Perform a cost-benefit analysis
Compute the benefits of the project (e.g. return on investment)
Compute the costs of the project - main factors:
● Data sources
● Data mining problem to be solved
● Available tools
● Expertise of the development team
Quantify the risk that the project fails:
● Knowledge not available
● Data not available
● Missing tools
Quality data & feature engineering
What are we going to learn today?
- What kind of data exists?
- How to prepare data?
- What is data balancing?
- How to apply data cleaning and feature scaling?
- What is feature selection?
, What kind of data exists?
- Structured data
- Unstructured data
- Semi-structured data
Structured data
Tabular data (rows and columns) which are very well defined
We know which columns there are and what kind of data they contain (the format is very
strict)
Often such data is stored in databases that represent the relationships between the data as
well. Questions about data can be answered by using a query language.
Unstructured data
The rawest form of data that can be any type of file.
Extracting value out of this shape of data is hard, since you need to extract structured
features from the data
For example, you might want to extract topics from movies.
Semi-structured data
This format is between structured and unstructured data
A consistent format is defined. However, the structure is not very strict. For example, it could
not be tabular or parts of the data may be missing.
Semi-structured data are often stored as files. However, some kinds of semi-structured data
can be stored in document oriented-databases. Such databases allow you to query the
sem-structured data
What are we going to learn today?
- What is machine learning?
- What are supervised and unsupervised machine learning?
- Which are the most common types of machine learning problems?
- Which are the basic steps of the CRoss Industry Standard Process for data mining
(CRISP-DM)?
Machine learning is the field of study that gives computers the ability to learn without being
explicitly programmed
Machine learning
Assume that you are iterating over and over again an exercise
What should be constant in your exercise?
- Learning! - machine learning applies strategies and algorithms, combined with data
and statistics
- Improving! - machine learning applies statistical indices to measure the overlap
between ML prediction and expected result
When you are doing it, it is human learning
When a machine does it, it is machine learning!
An example of supervised learning
Supervised learning - classification
Given a labelled dataset, the model learns to
predict new examples
An example of unsupervised learning
Unsupervised learning - clustering,
dimensionality reduction, anomaly detection
and novelty detection
Given a dataset, without labels, the model
learns to use to cluster/group similar data
,CRISP-DM process model
Business understanding in the CRISP-DM process
Determine business objectives and success criteria
Business objectives and measures to evaluate the results have to be established
Business objectives:
● What is the customer’s primary objective?
● Increase the number of loyal customers
● Selling more of a certain product
● Have a positive marketing campaign
,Business success criteria:
● Objective measure to establish success (e.g. return of investment)
Main steps in a data mining project
1. Define the goals:
Business and data mining experts together have to define the goals. For each goal a
measure must be defined to understand its success
2. Obtain the models:
Pre-process the data, apply data mining algorithms
3. Evaluate results
Use the pre-specified measures to evaluate the models
4. Deploy:
If the evaluation is successful, the model can be deployed
Costs & benefits
Perform a cost-benefit analysis
Compute the benefits of the project (e.g. return on investment)
Compute the costs of the project - main factors:
● Data sources
● Data mining problem to be solved
● Available tools
● Expertise of the development team
Quantify the risk that the project fails:
● Knowledge not available
● Data not available
● Missing tools
Quality data & feature engineering
What are we going to learn today?
- What kind of data exists?
- How to prepare data?
- What is data balancing?
- How to apply data cleaning and feature scaling?
- What is feature selection?
, What kind of data exists?
- Structured data
- Unstructured data
- Semi-structured data
Structured data
Tabular data (rows and columns) which are very well defined
We know which columns there are and what kind of data they contain (the format is very
strict)
Often such data is stored in databases that represent the relationships between the data as
well. Questions about data can be answered by using a query language.
Unstructured data
The rawest form of data that can be any type of file.
Extracting value out of this shape of data is hard, since you need to extract structured
features from the data
For example, you might want to extract topics from movies.
Semi-structured data
This format is between structured and unstructured data
A consistent format is defined. However, the structure is not very strict. For example, it could
not be tabular or parts of the data may be missing.
Semi-structured data are often stored as files. However, some kinds of semi-structured data
can be stored in document oriented-databases. Such databases allow you to query the
sem-structured data
