CSSE 386 Syllabus Schedule Worksheets/Quizzes Labs Resources

CSSE386 Data Mining Exam 1 Review Study Guide

This study guide covers key concepts, techniques, and example questions to prepare for Exam 1 in your Data Mining course CSSE386. Review each section carefully and check the solutions at the end.

---

Allowed Material

1. 1 sheet (2 sides) written notes
2. 2 printed handouts (see under Resources)

1. Data Processing and Cleaning

Key Topics:

• Handling missing data (e.g., imputation, removal).
• Removing duplicates and outliers.
• Standardization vs. normalization.

Example Questions:

1. True/False: Normalization rescales data to a range of [0, 1].
2. Fill in the Blanks: Imputing missing values is often preferred when the proportion of missing data is _____________. (hint: high or low)
3. Multiple Choice: Which method is most appropriate for imputing missing numerical data?
   • a) Mean imputation
   • b) Mode imputation
   • c) Deleting the row
   • d) Using a median filter

---

2. Data Visualization

Key Topics:

• Types of visualizations (e.g., scatter plots, histograms, box plots).
• Identifying trends and outliers through visualizations.

Example Questions:

1. Short Answer: What type of visualization is best suited to examine the distribution of numerical features?
2. True/False: Box plots are used to visualize the relationship between two categorical variables.

---

3. Unsupervised Learning

Key Topics:

• K-Means Clustering:
  • Choosing the number of clusters (“k”).
  • Interpret the results of the clustering.
• Hierarchical Clustering:
  • Dendrograms.

Example Questions:

1. True/False: In K-means, clusters are initialized randomly.
2. Multiple Choice: Which of the following measures is commonly used in K-means clustering?
   • a) Manhattan distance
   • b) Euclidean distance
   • c) Cosine similarity
   • d) Hamming distance
3. Short Answer: What does a dendrogram represent in hierarchical clustering?

---

4. Supervised Learning

Key Topics:

• Regression:
  • Linear regression and interpreting coefficients.
  • Mean Squared Error (MSE).
• Classification Methods:
  • Decision Trees: Splitting criteria and depth.
  • k-NN: Distance metrics and choosing \(k\).
  • SVM: Margin and kernel functions.
  • Naive Bayes: Assumptions and probabilities.

Example Questions:

1. True/False: Linear regression is sensitive to outliers.
2. Fill in the Blanks: SVM is effective for _____________ data (hint: low or high dimensional).
3. Multiple Choice: A typical value of k in k-NN is:
   • a) 0
   • b) 5
   • c) 2
   • d) 100

---

5. Confusion Matrix and Performance Metrics

Key Topics:

• Confusion matrix: TP, TN, FP, FN.
• Precision, recall, accuracy.
• Evaluation metrics for regression (e.g., MSE).

Example Questions:

1. Fill in the Blanks: Recall is defined as the fraction of _____________ that are correctly predicted.
2. True/False: The confusion matrix is applicable only to binary classification tasks.
3. Problem: Given the following confusion matrix, identify TP, TN, FP, FN:

   	Predicted Positive	Predicted Negative
   Actual Positive	50	10
   Actual Negative	5	35

4. Short Answer: Why might accuracy not be a reliable metric for imbalanced datasets?

---

6. Missing Data

Key Topics:

• Methods to handle missing data (mean, median, mode imputation).
• Impact of missing data on analysis.

Example Questions:

1. True/False: Imputation always improves the model performance.
2. Fill in the Blanks: Dropping columns with missing values is effective when the proportion of missing data is _____________.

---

General Tips for Exam Preparation

1. Review Lecture Notes: Focus on definitions, key formulas, and algorithms.
2. Practice Problems: Solve problems similar to the examples provided here.
3. Understand the Concepts: Avoid memorization; aim to understand the "why" behind each method.

Good luck with your exam preparation!

---

Answers

• 1. Data Processing and Cleaning
  1. True
  2. Low
  3. a) Mean imputation
• 2. Data Visualization
  1. Histograms
  2. False
• 3. Unsupervised Learning
  1. True
  2. b) Euclidean distance
  3. A dendrogram represents the hierarchical relationships between clusters.
• 4. Supervised Learning
  1. True
  2. high-dimensional
  3. b) 5
• 5. Confusion Matrix and Performance Metrics
  1. True Positives (TP) / (TP + FN)
  2. False
  3. TP:50, TN: 35, FN: 10, FP: 5
  4. Because it does not account for class imbalance.
• 6. Missing Data
  1. False
  2. High