1
ISYE 6501 Final Exam Practice Test (100 Questions)
Section 1: Classification & Machine Learning Models (Questions 1–25)
1. What type of learning does K-Nearest Neighbor (KNN) use?
A. Unsupervised learning
B. Supervised learning
C. Reinforcement learning
D. Semi-supervised learning
Answer: B (KNN is a supervised, non-parametric classification technique) .
2. What happens in KNN if k is set to 1?
A. The model overfits; the decision boundary is highly irregular
B. The model underfits
C. The model becomes parametric
D. The model cannot classify
Answer: A (Low k captures noise and high variance) .
3. Which of the following is a pro of K-Nearest Neighbor?
A. Works well with large datasets
B. No assumptions about data and easy to understand
C. Computationally inexpensive
D. Provides probability estimates directly
Answer: B (KNN is versatile and makes no assumptions, but is computationally expensive) .
4. What does SVM stand for?
A. Supervised Vector Machine
B. Support Vector Machine
C. Standard Vector Model
D. Statistical Vector Method
Answer: B (Support Vector Machine) .
5. What is the primary purpose of a Support Vector Machine?
A. To find a classifier with maximum separation or margin between two sets of points
B. To cluster data points
C. To forecast time series data
D. To reduce dimensionality
Answer: A (SVM finds the hyperplane with maximum margin) .
6. What are support vectors in SVM?
A. All data points
B. Data points closest to the decision boundary
C. Misclassified points
, 2
D. Outliers
Answer: B (Support vectors define the margin) .
7. What does a large lambda value in SVM signify?
A. The importance of a large margin outweighs avoiding mistakes and classifying known data
samples
B. The importance of minimizing mistakes outweighs having a large margin
C. The model is overfitting
D. The model is underfitting
Answer: A (As lambda gets larger, the margin term gets larger) .
8. What does a small lambda value (close to zero) in SVM signify?
A. The importance of a large margin outweighs avoiding mistakes
B. The importance of minimizing mistakes and classifying known data points outweighs
having a large margin
C. The model is perfectly accurate
D. The model is non-linear
Answer: B (As lambda drops towards zero, the error term dominates) .
9. Why is it important to scale data when using SVM?
A. To reduce computational time
B. Because if data has very different scales, a small change in one variable could swamp a
huge change in another when minimizing sum of squares of coefficients
C. Scaling is not important for SVM
D. To increase the margin
Answer: B (Different scales distort the objective function) .
10. What do kernel methods allow for in SVMs?
A. Nonlinear classifiers
B. Faster computation
C. Probability estimates
D. Feature selection
Answer: A (Kernel methods enable non-linear decision boundaries) .
11. Which model is an ensemble method?
A. Random Forest
B. K-Nearest Neighbor
C. Linear Regression
D. K-Means
Answer: A (Random Forest combines multiple decision trees) .
12. What does bagging reduce in ensemble models?
A. Bias
B. Variance
, 3
C. Accuracy
D. Margin
Answer: B (Bootstrap aggregating averages predictions to reduce variance) .
13. In decision trees, information gain is based on:
A. Entropy reduction
B. Variance reduction
C. Margin maximization
D. Distance minimization
Answer: A (Information gain measures purity increase after a split) .
14. Gini impurity is used for:
A. Clustering
B. Decision tree splits
C. SVM kernels
D. Time series forecasting
Answer: B (Gini measures misclassification probability) .
15. What is the main drawback of hierarchical clustering?
A. Requires specifying number of clusters in advance
B. High computational cost for large data (O(n³) or O(n²))
C. Cannot handle categorical data
D. Always produces different results
Answer: B (Hierarchical clustering is computationally intensive) .
16. What is a soft classifier and when is it used?
A. A classifier that always makes perfect predictions
B. A classifier that minimizes the number of mistakes when no line can separate all labeled
examples
C. A classifier that only works on soft data
D. A classifier that cannot handle noise
Answer: B (Soft classifiers trade off errors for a larger margin) .
17. What does a confusion matrix true positive rate measure?
A. Accuracy
B. Sensitivity or Recall (TP / (TP + FN))
C. Precision
D. Specificity
Answer: B (True positive rate equals sensitivity/recall) .
18. What is the purpose of an ROC curve?
A. To visualize the trade-off between True Positive Rate and False Positive Rate
B. To measure R-squared
C. To select variables
ISYE 6501 Final Exam Practice Test (100 Questions)
Section 1: Classification & Machine Learning Models (Questions 1–25)
1. What type of learning does K-Nearest Neighbor (KNN) use?
A. Unsupervised learning
B. Supervised learning
C. Reinforcement learning
D. Semi-supervised learning
Answer: B (KNN is a supervised, non-parametric classification technique) .
2. What happens in KNN if k is set to 1?
A. The model overfits; the decision boundary is highly irregular
B. The model underfits
C. The model becomes parametric
D. The model cannot classify
Answer: A (Low k captures noise and high variance) .
3. Which of the following is a pro of K-Nearest Neighbor?
A. Works well with large datasets
B. No assumptions about data and easy to understand
C. Computationally inexpensive
D. Provides probability estimates directly
Answer: B (KNN is versatile and makes no assumptions, but is computationally expensive) .
4. What does SVM stand for?
A. Supervised Vector Machine
B. Support Vector Machine
C. Standard Vector Model
D. Statistical Vector Method
Answer: B (Support Vector Machine) .
5. What is the primary purpose of a Support Vector Machine?
A. To find a classifier with maximum separation or margin between two sets of points
B. To cluster data points
C. To forecast time series data
D. To reduce dimensionality
Answer: A (SVM finds the hyperplane with maximum margin) .
6. What are support vectors in SVM?
A. All data points
B. Data points closest to the decision boundary
C. Misclassified points
, 2
D. Outliers
Answer: B (Support vectors define the margin) .
7. What does a large lambda value in SVM signify?
A. The importance of a large margin outweighs avoiding mistakes and classifying known data
samples
B. The importance of minimizing mistakes outweighs having a large margin
C. The model is overfitting
D. The model is underfitting
Answer: A (As lambda gets larger, the margin term gets larger) .
8. What does a small lambda value (close to zero) in SVM signify?
A. The importance of a large margin outweighs avoiding mistakes
B. The importance of minimizing mistakes and classifying known data points outweighs
having a large margin
C. The model is perfectly accurate
D. The model is non-linear
Answer: B (As lambda drops towards zero, the error term dominates) .
9. Why is it important to scale data when using SVM?
A. To reduce computational time
B. Because if data has very different scales, a small change in one variable could swamp a
huge change in another when minimizing sum of squares of coefficients
C. Scaling is not important for SVM
D. To increase the margin
Answer: B (Different scales distort the objective function) .
10. What do kernel methods allow for in SVMs?
A. Nonlinear classifiers
B. Faster computation
C. Probability estimates
D. Feature selection
Answer: A (Kernel methods enable non-linear decision boundaries) .
11. Which model is an ensemble method?
A. Random Forest
B. K-Nearest Neighbor
C. Linear Regression
D. K-Means
Answer: A (Random Forest combines multiple decision trees) .
12. What does bagging reduce in ensemble models?
A. Bias
B. Variance
, 3
C. Accuracy
D. Margin
Answer: B (Bootstrap aggregating averages predictions to reduce variance) .
13. In decision trees, information gain is based on:
A. Entropy reduction
B. Variance reduction
C. Margin maximization
D. Distance minimization
Answer: A (Information gain measures purity increase after a split) .
14. Gini impurity is used for:
A. Clustering
B. Decision tree splits
C. SVM kernels
D. Time series forecasting
Answer: B (Gini measures misclassification probability) .
15. What is the main drawback of hierarchical clustering?
A. Requires specifying number of clusters in advance
B. High computational cost for large data (O(n³) or O(n²))
C. Cannot handle categorical data
D. Always produces different results
Answer: B (Hierarchical clustering is computationally intensive) .
16. What is a soft classifier and when is it used?
A. A classifier that always makes perfect predictions
B. A classifier that minimizes the number of mistakes when no line can separate all labeled
examples
C. A classifier that only works on soft data
D. A classifier that cannot handle noise
Answer: B (Soft classifiers trade off errors for a larger margin) .
17. What does a confusion matrix true positive rate measure?
A. Accuracy
B. Sensitivity or Recall (TP / (TP + FN))
C. Precision
D. Specificity
Answer: B (True positive rate equals sensitivity/recall) .
18. What is the purpose of an ROC curve?
A. To visualize the trade-off between True Positive Rate and False Positive Rate
B. To measure R-squared
C. To select variables
