ISYE 6501 Final Exam 2026 | Questions and Answers | 100%
Correct - GT Actual Exam 2026/2027 | Complete Exam-Style
Questions with Detailed Rationales | 100% Verified | Pass
Guaranteed – A+ Graded
Classification and Regression Methods
Q1: A retail analyst uses k-Nearest Neighbors (k-NN) with k=1 to classify customer churn
based on a noisy dataset. What is the most likely outcome when evaluating this model on
new, unseen data?
A. The model will exhibit high bias and fail to capture underlying patterns.
B. The model will exhibit low bias and high variance, leading to severe overfitting.
[CORRECT]
C. The model will perfectly generalize because it memorizes the training data.
D. The model will underfit due to an overly simplified decision boundary.
Correct Answer: B
Rationale: The best answer is B because setting k=1 makes the model highly sensitive to
every noise point in the training set, resulting in a complex boundary with low bias but high
variance that fails to generalize.
Q2: A healthcare data scientist builds a logistic regression model to predict disease
presence. The coefficient for "age" is 0.05. How should this coefficient be interpreted in
terms of odds?
A. Each additional year of age increases the probability of the disease by exactly 5%.
B. Each additional year of age multiplies the odds of the disease by approximately 1.051.
[CORRECT]
C. Each additional year of age decreases the log-odds of the disease by 0.05.
D. The odds of the disease remain constant regardless of age.
Correct Answer: B
Rationale: This choice is correct because in logistic regression, exponentiating the coefficient
(e^0.05 ≈ 1.051) gives the odds ratio, meaning the odds increase by a factor of 1.051 for
each one-unit increase in the predictor.
Q3: A fraud detection team is building a Classification and Regression Tree (CART) and must
choose a splitting criterion. Why might they prefer Gini impurity over information gain
(entropy) for a massive dataset?
A. Gini impurity is computationally faster because it avoids calculating logarithmic functions.
[CORRECT]
B. Gini impurity always produces deeper, more accurate trees than entropy.
C. Entropy is strictly used for regression tasks, not classification.
D. Gini impurity naturally handles missing values better than entropy.
Correct Answer: A
,Rationale: This aligns with the principle that while both metrics measure node impurity
similarly, Gini impurity is computationally more efficient for large datasets since it does not
require computing logarithms.
Q4: An e-commerce company uses a Random Forest to predict delivery delays. How does
the "bagging" (bootstrap aggregating) process specifically help this model perform better
than a single decision tree?
A. It forces all trees to use the exact same subset of features to ensure consistency.
B. It reduces variance by training multiple trees on different random samples of the data
and averaging their predictions. [CORRECT]
C. It sequentially corrects the errors of previous trees to reduce bias.
D. It guarantees that the final model will have zero training error.
Correct Answer: B
Rationale: This matches the principle that bagging reduces the high variance inherent in
individual decision trees by averaging the predictions of multiple trees trained on
bootstrapped samples.
Q5: A manufacturing firm uses a Support Vector Machine (SVM) to classify defective parts.
What is the primary objective of the SVM algorithm during the training phase?
A. To minimize the number of misclassified points in the training set regardless of the
margin.
B. To find the hyperplane that maximizes the margin between the two closest points of
opposing classes. [CORRECT]
C. To calculate the probability that a new part belongs to the defective class.
D. To cluster the defective parts into distinct sub-categories before classification.
Correct Answer: B
Rationale: The best answer is B because the fundamental goal of a hard-margin SVM is to
identify the optimal separating hyperplane that maximizes the distance (margin) to the
nearest training data points (support vectors).
Q6: A real estate analyst notices that the variance of the residuals in their linear regression
model increases as the predicted house prices increase. What assumption of linear
regression has been violated?
A. Linearity of the relationship between predictors and the response.
B. Homoscedasticity, meaning the variance of the errors should be constant across all levels
of the predicted values. [CORRECT]
C. Independence of the predictor variables from one another.
D. Normality of the predictor variables.
Correct Answer: B
Rationale: This choice is correct because a funnel-shaped pattern in the residuals indicates
heteroscedasticity, violating the assumption that error variance remains constant across all
predicted values.
Q7: A genomics researcher has a dataset with 20,000 gene expressions (features) but only
100 patient samples. They want to perform feature selection to identify the most critical
genes. Which regularization method is best suited for this?
A. Ridge regression, because it shrinks all coefficients exactly to zero.
, B. Lasso regression, because its L1 penalty can shrink some coefficients exactly to zero,
performing automatic feature selection. [CORRECT]
C. Ordinary Least Squares, because it naturally ignores irrelevant features.
D. Elastic Net, because it strictly forbids any coefficients from becoming zero.
Correct Answer: B
Rationale: This aligns with the principle that Lasso regression applies an L1 penalty that
creates sparse models by forcing the coefficients of less important features to become
exactly zero.
Q8: A computer vision engineer applies k-NN to classify high-resolution images, treating
each pixel as a feature. The model performs terribly on the test set. What phenomenon is
primarily responsible for this failure?
A. The bias-variance tradeoff favoring high bias.
B. The curse of dimensionality, where distance metrics lose their meaningful distinction in
high-dimensional space. [CORRECT]
C. The lack of a sufficient training set to calculate the mean.
D. The inability of k-NN to handle continuous pixel values.
Correct Answer: B
Rationale: This matches the principle that in very high-dimensional spaces, the distance
between the closest and farthest points becomes nearly identical, rendering distance-based
algorithms like k-NN ineffective.
Q9: A marketing team uses logistic regression to predict if a user will click an ad. What is the
geometric shape of the decision boundary produced by a standard logistic regression model
without polynomial features?
A. A highly complex, non-linear curve.
B. A straight line or hyperplane in the feature space. [CORRECT]
C. A series of disjointed rectangular regions.
D. A circular boundary centered on the mean of the data.
Correct Answer: B
Rationale: The best answer is B because standard logistic regression is a linear classifier,
meaning its decision boundary in the original feature space is always a straight line (or
hyperplane in higher dimensions).
Q10: A credit scoring model using CART encounters several missing values in the "annual
income" feature. How does the CART algorithm typically handle these missing values during
the splitting process?
A. It deletes all rows containing any missing values before building the tree.
B. It uses surrogate splits, sending the observation down the branch of the best available
alternative feature. [CORRECT]
C. It imputes the missing values with the global mean of the dataset.
D. It creates a separate "missing" branch for every single node.
Correct Answer: B
Rationale: This choice is correct because CART handles missing data by identifying surrogate
variables that mimic the primary splitting variable, routing the observation based on the
surrogate's split.
Correct - GT Actual Exam 2026/2027 | Complete Exam-Style
Questions with Detailed Rationales | 100% Verified | Pass
Guaranteed – A+ Graded
Classification and Regression Methods
Q1: A retail analyst uses k-Nearest Neighbors (k-NN) with k=1 to classify customer churn
based on a noisy dataset. What is the most likely outcome when evaluating this model on
new, unseen data?
A. The model will exhibit high bias and fail to capture underlying patterns.
B. The model will exhibit low bias and high variance, leading to severe overfitting.
[CORRECT]
C. The model will perfectly generalize because it memorizes the training data.
D. The model will underfit due to an overly simplified decision boundary.
Correct Answer: B
Rationale: The best answer is B because setting k=1 makes the model highly sensitive to
every noise point in the training set, resulting in a complex boundary with low bias but high
variance that fails to generalize.
Q2: A healthcare data scientist builds a logistic regression model to predict disease
presence. The coefficient for "age" is 0.05. How should this coefficient be interpreted in
terms of odds?
A. Each additional year of age increases the probability of the disease by exactly 5%.
B. Each additional year of age multiplies the odds of the disease by approximately 1.051.
[CORRECT]
C. Each additional year of age decreases the log-odds of the disease by 0.05.
D. The odds of the disease remain constant regardless of age.
Correct Answer: B
Rationale: This choice is correct because in logistic regression, exponentiating the coefficient
(e^0.05 ≈ 1.051) gives the odds ratio, meaning the odds increase by a factor of 1.051 for
each one-unit increase in the predictor.
Q3: A fraud detection team is building a Classification and Regression Tree (CART) and must
choose a splitting criterion. Why might they prefer Gini impurity over information gain
(entropy) for a massive dataset?
A. Gini impurity is computationally faster because it avoids calculating logarithmic functions.
[CORRECT]
B. Gini impurity always produces deeper, more accurate trees than entropy.
C. Entropy is strictly used for regression tasks, not classification.
D. Gini impurity naturally handles missing values better than entropy.
Correct Answer: A
,Rationale: This aligns with the principle that while both metrics measure node impurity
similarly, Gini impurity is computationally more efficient for large datasets since it does not
require computing logarithms.
Q4: An e-commerce company uses a Random Forest to predict delivery delays. How does
the "bagging" (bootstrap aggregating) process specifically help this model perform better
than a single decision tree?
A. It forces all trees to use the exact same subset of features to ensure consistency.
B. It reduces variance by training multiple trees on different random samples of the data
and averaging their predictions. [CORRECT]
C. It sequentially corrects the errors of previous trees to reduce bias.
D. It guarantees that the final model will have zero training error.
Correct Answer: B
Rationale: This matches the principle that bagging reduces the high variance inherent in
individual decision trees by averaging the predictions of multiple trees trained on
bootstrapped samples.
Q5: A manufacturing firm uses a Support Vector Machine (SVM) to classify defective parts.
What is the primary objective of the SVM algorithm during the training phase?
A. To minimize the number of misclassified points in the training set regardless of the
margin.
B. To find the hyperplane that maximizes the margin between the two closest points of
opposing classes. [CORRECT]
C. To calculate the probability that a new part belongs to the defective class.
D. To cluster the defective parts into distinct sub-categories before classification.
Correct Answer: B
Rationale: The best answer is B because the fundamental goal of a hard-margin SVM is to
identify the optimal separating hyperplane that maximizes the distance (margin) to the
nearest training data points (support vectors).
Q6: A real estate analyst notices that the variance of the residuals in their linear regression
model increases as the predicted house prices increase. What assumption of linear
regression has been violated?
A. Linearity of the relationship between predictors and the response.
B. Homoscedasticity, meaning the variance of the errors should be constant across all levels
of the predicted values. [CORRECT]
C. Independence of the predictor variables from one another.
D. Normality of the predictor variables.
Correct Answer: B
Rationale: This choice is correct because a funnel-shaped pattern in the residuals indicates
heteroscedasticity, violating the assumption that error variance remains constant across all
predicted values.
Q7: A genomics researcher has a dataset with 20,000 gene expressions (features) but only
100 patient samples. They want to perform feature selection to identify the most critical
genes. Which regularization method is best suited for this?
A. Ridge regression, because it shrinks all coefficients exactly to zero.
, B. Lasso regression, because its L1 penalty can shrink some coefficients exactly to zero,
performing automatic feature selection. [CORRECT]
C. Ordinary Least Squares, because it naturally ignores irrelevant features.
D. Elastic Net, because it strictly forbids any coefficients from becoming zero.
Correct Answer: B
Rationale: This aligns with the principle that Lasso regression applies an L1 penalty that
creates sparse models by forcing the coefficients of less important features to become
exactly zero.
Q8: A computer vision engineer applies k-NN to classify high-resolution images, treating
each pixel as a feature. The model performs terribly on the test set. What phenomenon is
primarily responsible for this failure?
A. The bias-variance tradeoff favoring high bias.
B. The curse of dimensionality, where distance metrics lose their meaningful distinction in
high-dimensional space. [CORRECT]
C. The lack of a sufficient training set to calculate the mean.
D. The inability of k-NN to handle continuous pixel values.
Correct Answer: B
Rationale: This matches the principle that in very high-dimensional spaces, the distance
between the closest and farthest points becomes nearly identical, rendering distance-based
algorithms like k-NN ineffective.
Q9: A marketing team uses logistic regression to predict if a user will click an ad. What is the
geometric shape of the decision boundary produced by a standard logistic regression model
without polynomial features?
A. A highly complex, non-linear curve.
B. A straight line or hyperplane in the feature space. [CORRECT]
C. A series of disjointed rectangular regions.
D. A circular boundary centered on the mean of the data.
Correct Answer: B
Rationale: The best answer is B because standard logistic regression is a linear classifier,
meaning its decision boundary in the original feature space is always a straight line (or
hyperplane in higher dimensions).
Q10: A credit scoring model using CART encounters several missing values in the "annual
income" feature. How does the CART algorithm typically handle these missing values during
the splitting process?
A. It deletes all rows containing any missing values before building the tree.
B. It uses surrogate splits, sending the observation down the branch of the best available
alternative feature. [CORRECT]
C. It imputes the missing values with the global mean of the dataset.
D. It creates a separate "missing" branch for every single node.
Correct Answer: B
Rationale: This choice is correct because CART handles missing data by identifying surrogate
variables that mimic the primary splitting variable, routing the observation based on the
surrogate's split.
