Week 1
The provided text appears to be an excerpt or outline of a course syllabus or lecture notes
on the topic of data mining. Here's a summary of the key points covered in this material:
Course Overview:
- The course is structured to include theoretical lectures and practical sessions, with a focus
on both theoretical knowledge and hands-on coding skills.
- Course materials, including lecture content, will be published weekly before the theory
lecture.
- Evaluation for the course will be based on a final exam, which is written, on-campus, and
closed-book. The exam consists of multiple-choice questions carrying equal weight.
Remark on Final Exam:
- The final exam will include code-related questions, particularly in Python.
- Weekly quizzes with multiple-choice questions resembling those in the final exam will be
provided on the Canvas platform. These quizzes do not count towards the final grade but are
encouraged for practice.
Additional Information:
- Correct answers and justifications for quizzes will be released on Fridays.
- The course will include reading material consisting of selected book chapters, which is
optional but highly recommended to enhance understanding of theoretical concepts
discussed in lectures.
Getting Started: Pattern Classification:
- The course introduces the concept of pattern classification, where numerical variables
(features) are used to predict outcomes (decision classes). This is a multi-class problem.
- The goal in pattern classification is to build models that can generalize well beyond
historical training data.
Dealing with New Instances:
- When encountering new instances, the course will cover how to apply the trained model to
make predictions.
- The course will discuss topics like handling missing values, computing
correlations/associations between features, and encoding categorical features. These are
part of pre-processing and exploratory data analysis steps.
Handling Missing Values:
- Missing values in data can arise from various reasons, and it's crucial to address them
before building machine learning or data mining models.
- Strategies for handling missing values include removing the feature, removing instances, or
imputing missing values using techniques such as mean, median, mode, or machine learning
models.
Autoencoders for Imputing Missing Values:
- Autoencoders, which are deep neural networks with encoder and decoder blocks, can be
used for imputing missing values in data through unsupervised learning.
, Feature Scaling:
- Feature scaling techniques like normalization and standardization are discussed to bring
features to similar scales, preventing issues with extreme values.
Feature Interaction:
- Methods for measuring correlation between numerical features and association between
categorical features are discussed. Pearson's correlation coefficient is introduced for
numerical features, and the chi-squared measure is mentioned for categorical features.
Encoding Categorical Features:
- Strategies for encoding categorical features, including label encoding for ordinal
relationships and one-hot encoding for nominal features, are explained.
Dealing with Class Imbalance:
- Class imbalance in classification problems is addressed, and strategies like random instance
selection, creating synthetic instances (SMOTE), and associated considerations are discussed.
Course Focus:
- The course is primarily oriented toward data mining for business and governance
applications.
This material outlines the structure and content of the course, highlighting the importance
of theoretical knowledge and practical skills in data mining, along with specific techniques
and strategies used in data preprocessing, feature handling, and class imbalance
management.
Week 2
The material you provided seems to be from a course on pattern classification and data
mining for business and governance, possibly a lecture or presentation by Dr. Gonzalo
Nápoles. Here's a summary of the key points covered in this material:
1. Classification Problem : The material discusses a classification problem where the goal is
to predict outcomes based on four categorical features. This is a binary classification
problem with two possible outcomes or decision classes.
2. Data : The provided data includes features like Outlook, Temperature, Humidity, Windy,
and Play, along with corresponding outcomes for training the classification model.
3. Approaches to Classification :
- Rule-Based Learning : This approach involves creating a set of rules based on features
and their values to make predictions. Decision trees are commonly used for this purpose.
- Bayesian Learning : Bayesian learning utilizes probabilities to make predictions, assuming
independence among features. Naïve Bayes is a popular algorithm in this category.
- Lazy Learning : Lazy learning relies on similarity between instances to make predictions.
The k-Nearest Neighbors (k-NN) algorithm is an example.
