CSSE386 Data Mining Exam 2 Review Study Guide

This study guide covers key concepts, techniques, and example questions to prepare for Exam 2 in your Data Mining course CSSE386. Review each section carefully and check the solutions at the end.

Information

1. Dimensionality Reduction

Key Topics:

Example Questions:

  1. Multiple Choice: What is the primary goal of dimensionality reduction?
    1. Increase the number of features
    2. Reduce noise and computational complexity while preserving essential structure
    3. Overfit the data
    4. Separate data into clusters
  2. Multiple Choice: Which statement best differentiates feature reduction from feature selection?
    1. Feature reduction transforms the original features into a new, lower-dimensional space, whereas feature selection chooses a subset of the original features without transformation.
    2. Feature reduction chooses a subset of the original features, whereas feature selection transforms the data into a new feature space.
    3. Both methods perform the same operation and the terms are interchangeable.

    Topic 2: Principal Component Analysis (PCA)

    Key Topics:

    • Understanding the PCA technique for dimensionality reduction
    • Understanding the covariance matrix
    • Choosing principal components
    • Interpreting principal components from plotted results

    Example Questions:

    1. True/False: PCA is a supervised learning method
    2. Multiple Choice: In PCA, the eigenvectors corresponding to the largest eigenvalues represent:
      1. Noise directions
      2. Directions with maximum variance
      3. Random directions
      4. Directions with minimum variance
    3. Multiple Choice: Which statement best describes the elements of a covariance matrix?
      1. Diagonal elements represent the mean of each feature, while off-diagonals represent the correlation between features
      2. Diagonal elements represent the variances of each feature, while off-diagonals represent the covariances between pairs of features
      3. Diagonal elements represent the standard deviations of each feature, while off-diagonals are always zero
      4. Diagonal elements represent the eigenvalues of the data, while off-diagonals represent the eigenvectors

    Topic 3: Linear Discriminant Analysis (LDA)

    Key Topics:

    • Understanding LDA techniques
    • Recognizing differences between LDA and PCA

    Example Questions:

    1. Multiple Choice: Which method is specifically designed to maximize the separability between different classes?
      1. PCA
      2. LDA
      3. k-means clustering
      4. Support Vector Machines
    2. Multiple Choice: Which statement best describes Linear Discriminant Analysis (LDA)?
      1. It finds a projection that maximizes the overall variance in the data
      2. It projects data onto a subspace that maximizes the ratio of between-class variance to within-class variance
      3. It employs a non-linear mapping to reduce dimensionality
      4. It ignores class labels during the dimensionality reduction process
    3. Multiple Choice: Which of the following best distinguishes PCA from LDA?
      1. PCA is unsupervised whereas LDA is supervised
      2. PCA maximizes between-class variance while LDA maximizes overall variance
      3. PCA uses class labels in its computation, while LDA does not
      4. Both methods are identical in their approach

    Topic 4: Recommendation Systems

    Key Topics:

    • Collaborative Filtering vs Content-Based Filtering
    • Memory-based vs Model-based techniques
    • Explicit vs Implicit information
    • Cosine similarity (pros/cons)
    • Matrix Factorization (pros/cons)

    Example Questions:

    1. Multiple Choice: In recommendation systems, which approach primarily relies on analyzing item features and content to suggest items to users?
      1. Collaborative filtering
      2. Content-based filtering
      3. Hybrid filtering
      4. Matrix factorization
    2. Multiple Choice: A potential drawback of content-based filtering:
      1. Requires large amounts of user interaction data
      2. May overspecialize recommendations, limiting diversity ("filter bubble")
      3. Cannot incorporate item features
      4. Is computationally intensive in high-dimensional spaces
    3. Multiple Choice: matrix factorization techniques are primarily used to:
      1. Cluster users into distinct groups
      2. Decompose the user–item interaction matrix to identify latent factors
      3. Normalize user ratings
      4. Remove outliers from the dataset

    General Tips for Exam Preparation

    1. Review Lecture Notes: Focus on definitions, key formulas, and algorithms
    2. Understand the Concepts: Avoid memorization; aim to understand the "why" behind each method

    Good luck with your exam preparation!

    Answer Key with Explanations

    Topic 1: Dimensionality Reduction

    1. Answer: b. Explanation: The goal is to reduce noise and computational complexity while preserving the essential structure of the data
    2. Answer: a. Explanation: Feature reduction (e.g., via PCA) transforms the original features into a new, lower-dimensional space, whereas feature selection picks a subset of the original features without altering their form

    Topic 2: Principal Component Analysis (PCA)

    1. Answer: False. Explanation: PCA is an unsupervised method; it does not use label information
    2. Answer: b. Explanation: The eigenvectors corresponding to the largest eigenvalues indicate the directions of maximum variance in the data.
    3. Answer: b. Explanation: Diagonal Elements are the variances of the individual features, off-Diagonal Elements represent the covariances between different features

    Topic 3: Linear Discriminant Analysis (LDA)

    1. Answer: b. Explanation: LDA is specifically designed to maximize the separability between different classes by using class label information.
    2. Answer: b. Explanation: LDA projects data onto a subspace that maximizes the ratio of between-class variance to within-class variance, thereby enhancing class separability
    3. Answer: a. Explanation: The key distinction is that PCA is unsupervised (ignores class labels), whereas LDA is a supervised technique that uses class labels to find the optimal projection for separation

    Topic 4: Recommendation Systems

    1. Answer: b. Explanation: Content-based filtering relies on analyzing the features of items (such as keywords or attributes) to make recommendations
    2. Answer: b. Explanation: A drawback of content-based filtering is that it may overspecialize recommendations, limiting the diversity of suggestions
    3. Answer: b. Explanation: Matrix factorization techniques decompose the sparce user–item interaction matrix into 2 small matrices (user and items) to uncover latent factors