ISYE 6501 Final Exam Actual Q&A – Complete Solution 2026/2027 | Georgia Tech
Analytics Modeling with Detailed Rationales | 100% Verified | Pass Guaranteed – A+
Graded
Section A: Classification Models - Logistic Regression, SVM, & K-NN
(12 Questions)
Q1: A logistic regression model for loan default prediction outputs a coefficient of 0.8
for the "debt-to-income ratio" variable. Which interpretation is correct?
A. A one-unit increase in debt-to-income ratio increases the probability of default by 0.8
B. A one-unit increase in debt-to-income ratio multiplies the odds of default by e^0.8 ≈
2.23 [CORRECT]
C. A one-unit increase in debt-to-income ratio increases the default probability by 80%
D. The coefficient directly represents the probability change
Correct Answer: B
Rationale: In logistic regression, coefficients represent log-odds changes. A coefficient
β means a one-unit increase multiplies the odds by e^β. e^0.8 ≈ 2.23. Coefficients do
not directly represent probability changes (A, D) or percentage probability increases (C).
Q2: A support vector machine (SVM) with a linear kernel is trained on a dataset. The
model achieves 95% accuracy on training data but 72% on test data. Which adjustment
is MOST likely to improve generalization?
,A. Decrease the regularization parameter C (allow more misclassification on training
data) [CORRECT]
B. Increase the regularization parameter C (penalize misclassification more heavily)
C. Switch to an RBF kernel with default parameters
D. Add more features without regularization
Correct Answer: A
Rationale: The large gap between training (95%) and test (72%) accuracy indicates
overfitting. Decreasing C (the regularization parameter) allows more training
misclassification, increasing the margin and improving generalization. Increasing C (B)
worsens overfitting. RBF kernel (C) may overfit more. Adding features (D) without
regularization worsens overfitting.
Q3: In a K-Nearest Neighbors (K-NN) classifier, which issue is MOST likely to occur if the
features are on vastly different scales (e.g., income in dollars vs. age in years)?
A. The algorithm will converge faster
B. Features with larger scales will dominate distance calculations [CORRECT]
C. K-NN is inherently scale-invariant
D. The choice of K becomes irrelevant
Correct Answer: B
Rationale: K-NN uses distance metrics (Euclidean, Manhattan) that are sensitive to
feature scales. Without standardization, features with larger ranges (e.g., income in
, thousands) dominate over features with smaller ranges (e.g., age in years). K-NN is NOT
scale-invariant (C). Scaling does not affect convergence speed (A) or K relevance (D).
Q4: A confusion matrix for a medical diagnostic test shows: True Positives = 85, False
Positives = 15, True Negatives = 90, False Negatives = 10. What is the sensitivity
(recall)?
A. 85%
B. 89.5% [CORRECT]
C. 93.3%
D. 85/175
Correct Answer: B
Rationale: Sensitivity = TP / (TP + FN) = 85 / (85 + 10) = 85/95 = 0.8947 ≈ 89.5%. Option
A is just TP. Option C is precision (TP/(TP+FP) = 85/100). Option D uses wrong
denominator.
Q5: Which kernel function in SVM is MOST appropriate when the decision boundary is
known to be nonlinear and the data has complex, irregular patterns?
A. Linear kernel
B. Polynomial kernel
C. Radial Basis Function (RBF) kernel [CORRECT]
D. No kernel (linear SVM only)
Analytics Modeling with Detailed Rationales | 100% Verified | Pass Guaranteed – A+
Graded
Section A: Classification Models - Logistic Regression, SVM, & K-NN
(12 Questions)
Q1: A logistic regression model for loan default prediction outputs a coefficient of 0.8
for the "debt-to-income ratio" variable. Which interpretation is correct?
A. A one-unit increase in debt-to-income ratio increases the probability of default by 0.8
B. A one-unit increase in debt-to-income ratio multiplies the odds of default by e^0.8 ≈
2.23 [CORRECT]
C. A one-unit increase in debt-to-income ratio increases the default probability by 80%
D. The coefficient directly represents the probability change
Correct Answer: B
Rationale: In logistic regression, coefficients represent log-odds changes. A coefficient
β means a one-unit increase multiplies the odds by e^β. e^0.8 ≈ 2.23. Coefficients do
not directly represent probability changes (A, D) or percentage probability increases (C).
Q2: A support vector machine (SVM) with a linear kernel is trained on a dataset. The
model achieves 95% accuracy on training data but 72% on test data. Which adjustment
is MOST likely to improve generalization?
,A. Decrease the regularization parameter C (allow more misclassification on training
data) [CORRECT]
B. Increase the regularization parameter C (penalize misclassification more heavily)
C. Switch to an RBF kernel with default parameters
D. Add more features without regularization
Correct Answer: A
Rationale: The large gap between training (95%) and test (72%) accuracy indicates
overfitting. Decreasing C (the regularization parameter) allows more training
misclassification, increasing the margin and improving generalization. Increasing C (B)
worsens overfitting. RBF kernel (C) may overfit more. Adding features (D) without
regularization worsens overfitting.
Q3: In a K-Nearest Neighbors (K-NN) classifier, which issue is MOST likely to occur if the
features are on vastly different scales (e.g., income in dollars vs. age in years)?
A. The algorithm will converge faster
B. Features with larger scales will dominate distance calculations [CORRECT]
C. K-NN is inherently scale-invariant
D. The choice of K becomes irrelevant
Correct Answer: B
Rationale: K-NN uses distance metrics (Euclidean, Manhattan) that are sensitive to
feature scales. Without standardization, features with larger ranges (e.g., income in
, thousands) dominate over features with smaller ranges (e.g., age in years). K-NN is NOT
scale-invariant (C). Scaling does not affect convergence speed (A) or K relevance (D).
Q4: A confusion matrix for a medical diagnostic test shows: True Positives = 85, False
Positives = 15, True Negatives = 90, False Negatives = 10. What is the sensitivity
(recall)?
A. 85%
B. 89.5% [CORRECT]
C. 93.3%
D. 85/175
Correct Answer: B
Rationale: Sensitivity = TP / (TP + FN) = 85 / (85 + 10) = 85/95 = 0.8947 ≈ 89.5%. Option
A is just TP. Option C is precision (TP/(TP+FP) = 85/100). Option D uses wrong
denominator.
Q5: Which kernel function in SVM is MOST appropriate when the decision boundary is
known to be nonlinear and the data has complex, irregular patterns?
A. Linear kernel
B. Polynomial kernel
C. Radial Basis Function (RBF) kernel [CORRECT]
D. No kernel (linear SVM only)
