ANALYTICS MODELING
ISYE 6501 Final Quiz 2026/2027
A+
Complete Blueprint Coverage
A+ 5 100%
QUESTIONS VERIFIED EXAM DOMAINS COVERED RATIONALES INCLUDED
CATEGORIES
Classification, Clustering, and Validation
Time Series, Change Detection, and Regression
Tree-Based Models, Variable Selection, and Optimization
Data Preparation, Outliers, PCA, and Missing Data
Simulation, Probability Models, DOE, and Combining Models
STUVIAACTUALEXAM
, SECTION 1: CLASSIFICATION, CLUSTERING, AND VALIDATION
Q1
A credit-card fraud team trains a linear SVM on a data set in which a small fraction of fraudulent transactions are not linearly
separable from legitimate ones. Increasing the soft-margin penalty parameter C forces the model to classify almost every training
point correctly. Which consequence is most likely?
A. The model overfits the training data, producing a complex boundary that may perform poorly on unseen fraud patterns.
B. The decision boundary becomes smoother and generalizes better to new transactions.
C. The SVM automatically switches to an unsupervised clustering mode.
D. All kernel functions become equivalent and the dual problem disappears.
Correct Answer: A
Rationale:
A large C places heavy penalty on margin violations, encouraging the optimizer to fit the training points tightly. When the data are not linearly
separable this produces an overfit boundary that captures noise rather than the true fraud signal.
Q2
An analyst is building a k-nearest-neighbors classifier for customer churn. The feature set mixes dollar amounts, ages, and binary
indicators. After trying several values of k, performance on a hold-out set remains poor. Which preprocessing step is most likely to
improve distance-based classification?
A. Removing all continuous features so that only binary indicators remain.
B. Standardizing or normalizing the numeric features so that scale differences do not dominate Euclidean distance calculations.
C. Increasing k to the full size of the training set so that every point is used.
D. Replacing the distance metric with a random projection that ignores feature magnitudes.
Correct Answer: B
Rationale:
k-NN relies on distance. When features have incompatible scales, large-magnitude variables dominate the distance calculation. Standardization
places features on a common scale and typically improves neighbor selection.
Q3
A marketing team runs k-means with k = 5 on customer spending data. After 20 random initializations the within-cluster sum of
squares still varies by more than 30 % across runs. Which statement best explains the observed behavior?
A. k-means is guaranteed to reach the global optimum from any starting point.
B. The algorithm is unsupervised and therefore has no objective function to optimize.
C. k-means optimizes a non-convex objective and can converge to different local minima depending on initialization.
D. Variation across runs indicates that the data contain no cluster structure whatsoever.
Correct Answer: C
Rationale:
The k-means objective is non-convex. Different random starts can lead to different local minima, producing different partitions and different values of
the within-cluster sum of squares.
Q4
An operations analyst plots the within-cluster sum of squares for k-means as k ranges from 1 to 15 and observes a sharp “elbow” at
k = 4 followed by only gradual improvement. Which decision is most consistent with the elbow heuristic?
A. Select k = 15 because larger k always yields lower within-cluster variance.
B. Ignore the plot and choose k equal to the number of features.
C. Conclude that clustering is inappropriate for any data set that shows an elbow.
D. Prefer k = 4 as a pragmatic trade-off between compactness and simplicity, then validate with domain knowledge or silhouette scores.
Correct Answer: D
Rationale:
The elbow method looks for the point of diminishing returns in the plot of within-cluster variance versus k. Selecting that k balances fit and
parsimony; further validation is still recommended.
STUVIAACTUALEXAM ISYE 6501 Final Quiz 2026/2027
ISYE 6501 Final Quiz 2026/2027
A+
Complete Blueprint Coverage
A+ 5 100%
QUESTIONS VERIFIED EXAM DOMAINS COVERED RATIONALES INCLUDED
CATEGORIES
Classification, Clustering, and Validation
Time Series, Change Detection, and Regression
Tree-Based Models, Variable Selection, and Optimization
Data Preparation, Outliers, PCA, and Missing Data
Simulation, Probability Models, DOE, and Combining Models
STUVIAACTUALEXAM
, SECTION 1: CLASSIFICATION, CLUSTERING, AND VALIDATION
Q1
A credit-card fraud team trains a linear SVM on a data set in which a small fraction of fraudulent transactions are not linearly
separable from legitimate ones. Increasing the soft-margin penalty parameter C forces the model to classify almost every training
point correctly. Which consequence is most likely?
A. The model overfits the training data, producing a complex boundary that may perform poorly on unseen fraud patterns.
B. The decision boundary becomes smoother and generalizes better to new transactions.
C. The SVM automatically switches to an unsupervised clustering mode.
D. All kernel functions become equivalent and the dual problem disappears.
Correct Answer: A
Rationale:
A large C places heavy penalty on margin violations, encouraging the optimizer to fit the training points tightly. When the data are not linearly
separable this produces an overfit boundary that captures noise rather than the true fraud signal.
Q2
An analyst is building a k-nearest-neighbors classifier for customer churn. The feature set mixes dollar amounts, ages, and binary
indicators. After trying several values of k, performance on a hold-out set remains poor. Which preprocessing step is most likely to
improve distance-based classification?
A. Removing all continuous features so that only binary indicators remain.
B. Standardizing or normalizing the numeric features so that scale differences do not dominate Euclidean distance calculations.
C. Increasing k to the full size of the training set so that every point is used.
D. Replacing the distance metric with a random projection that ignores feature magnitudes.
Correct Answer: B
Rationale:
k-NN relies on distance. When features have incompatible scales, large-magnitude variables dominate the distance calculation. Standardization
places features on a common scale and typically improves neighbor selection.
Q3
A marketing team runs k-means with k = 5 on customer spending data. After 20 random initializations the within-cluster sum of
squares still varies by more than 30 % across runs. Which statement best explains the observed behavior?
A. k-means is guaranteed to reach the global optimum from any starting point.
B. The algorithm is unsupervised and therefore has no objective function to optimize.
C. k-means optimizes a non-convex objective and can converge to different local minima depending on initialization.
D. Variation across runs indicates that the data contain no cluster structure whatsoever.
Correct Answer: C
Rationale:
The k-means objective is non-convex. Different random starts can lead to different local minima, producing different partitions and different values of
the within-cluster sum of squares.
Q4
An operations analyst plots the within-cluster sum of squares for k-means as k ranges from 1 to 15 and observes a sharp “elbow” at
k = 4 followed by only gradual improvement. Which decision is most consistent with the elbow heuristic?
A. Select k = 15 because larger k always yields lower within-cluster variance.
B. Ignore the plot and choose k equal to the number of features.
C. Conclude that clustering is inappropriate for any data set that shows an elbow.
D. Prefer k = 4 as a pragmatic trade-off between compactness and simplicity, then validate with domain knowledge or silhouette scores.
Correct Answer: D
Rationale:
The elbow method looks for the point of diminishing returns in the plot of within-cluster variance versus k. Selecting that k balances fit and
parsimony; further validation is still recommended.
STUVIAACTUALEXAM ISYE 6501 Final Quiz 2026/2027
