ISYE 6501 INTRODUCTION TO
ANALYTICS MODELING FINAL EXAM
PREP QUESTIONS AND ANSWERS
1. In Support Vector Machines (SVM), what is the primary purpose of the ‘C’ parameter?
A. To control the trade-off between maximizing the margin and minimizing classification
errors.
B. To determine the kernel type used for mapping data to higher dimensions.
C. To define the number of support vectors allowed in the final model.
D. To set the learning rate for the gradient descent optimization process.
Answer: A
Conceptual Explanation: The C parameter acts as a regularization parameter; a large C
penalizes misclassifications heavily (narrow margin), while a small C allows more
misclassifications for a wider margin (potentially better generalization).
2. Which of the following scenarios is best suited for using a CUSUM (Cumulative Sum)
control chart?
A. Classifying images into different categories of animals.
B. Predicting the exact price of a stock ten years into the future.
,C. Detecting a small but persistent shift in the mean of a process over time.
D. Clustering customers based on their geographical location.
Answer: C
Conceptual Explanation: CUSUM is specifically designed for change-point detection,
making it very effective at identifying when a process mean has shifted from its target.
3. When performing k-fold cross-validation, what is the primary risk of choosing a value of ‘k’
that is too small (e.g., k=2)?
A. The model will likely overfit the training data significantly.
B. The variance of the performance estimate will be zero.
C. The computational cost will become prohibitively expensive.
D. The estimate of the model’s performance will have high bias.
Answer: D
Conceptual Explanation: Lower k (like 2-fold) uses less data for training in each fold,
leading to an overestimation of the error (high bias), whereas higher k (like Leave-One-
Out) has lower bias but higher variance.
4. In the context of k-Nearest Neighbors (k-NN), why is it critical to scale/normalize features
before modeling?
A. To ensure the algorithm can handle categorical variables directly.
B. To transform the data into a normal distribution for statistical testing.
, C. To reduce the number of neighbors required for an accurate prediction.
D. Because k-NN relies on distance metrics which are sensitive to the magnitude of
features.
Answer: D
Conceptual Explanation: k-NN uses distance (like Euclidean) to find neighbors. Features
with larger scales will dominate the distance calculation if not normalized.
5. Which component of the Holt-Winters method accounts for recurring patterns over a fixed
period?
A. The Level component
B. The Trend component
C. The Seasonality component
D. The Error component
Answer: C
Conceptual Explanation: The triple exponential smoothing (Holt-Winters) includes Level,
Trend, and Seasonality. Seasonality specifically handles cyclic patterns (e.g., daily, weekly,
or monthly).
6. In Principal Component Analysis (PCA), what does the first principal component represent?
A. The variable with the highest mean in the dataset.
B. The direction in the data that captures the maximum variance.
ANALYTICS MODELING FINAL EXAM
PREP QUESTIONS AND ANSWERS
1. In Support Vector Machines (SVM), what is the primary purpose of the ‘C’ parameter?
A. To control the trade-off between maximizing the margin and minimizing classification
errors.
B. To determine the kernel type used for mapping data to higher dimensions.
C. To define the number of support vectors allowed in the final model.
D. To set the learning rate for the gradient descent optimization process.
Answer: A
Conceptual Explanation: The C parameter acts as a regularization parameter; a large C
penalizes misclassifications heavily (narrow margin), while a small C allows more
misclassifications for a wider margin (potentially better generalization).
2. Which of the following scenarios is best suited for using a CUSUM (Cumulative Sum)
control chart?
A. Classifying images into different categories of animals.
B. Predicting the exact price of a stock ten years into the future.
,C. Detecting a small but persistent shift in the mean of a process over time.
D. Clustering customers based on their geographical location.
Answer: C
Conceptual Explanation: CUSUM is specifically designed for change-point detection,
making it very effective at identifying when a process mean has shifted from its target.
3. When performing k-fold cross-validation, what is the primary risk of choosing a value of ‘k’
that is too small (e.g., k=2)?
A. The model will likely overfit the training data significantly.
B. The variance of the performance estimate will be zero.
C. The computational cost will become prohibitively expensive.
D. The estimate of the model’s performance will have high bias.
Answer: D
Conceptual Explanation: Lower k (like 2-fold) uses less data for training in each fold,
leading to an overestimation of the error (high bias), whereas higher k (like Leave-One-
Out) has lower bias but higher variance.
4. In the context of k-Nearest Neighbors (k-NN), why is it critical to scale/normalize features
before modeling?
A. To ensure the algorithm can handle categorical variables directly.
B. To transform the data into a normal distribution for statistical testing.
, C. To reduce the number of neighbors required for an accurate prediction.
D. Because k-NN relies on distance metrics which are sensitive to the magnitude of
features.
Answer: D
Conceptual Explanation: k-NN uses distance (like Euclidean) to find neighbors. Features
with larger scales will dominate the distance calculation if not normalized.
5. Which component of the Holt-Winters method accounts for recurring patterns over a fixed
period?
A. The Level component
B. The Trend component
C. The Seasonality component
D. The Error component
Answer: C
Conceptual Explanation: The triple exponential smoothing (Holt-Winters) includes Level,
Trend, and Seasonality. Seasonality specifically handles cyclic patterns (e.g., daily, weekly,
or monthly).
6. In Principal Component Analysis (PCA), what does the first principal component represent?
A. The variable with the highest mean in the dataset.
B. The direction in the data that captures the maximum variance.
