ISYE 6501 Final Quiz Actual Exam 2026/2027: Complete
Exam-Style Questions with Detailed Rationales | 100%
Verified | Pass Guaranteed – A+ Graded
TABLE OF CONTENTS
Section 1 | Classification Models | Q1 – Q10
Section 2 | Clustering and Dimensionality Reduction | Q11 – Q20
Section 3 | Time Series and Forecasting | Q21 – Q30
Section 4 | Validation and Model Selection | Q31 – Q40
Section 5 | Optimization and Linear Programming | Q41 – Q50
Instructions: Choose the single best answer. Pass: 80% in 90 minutes.
══════════════════════════════════════
SECTION 1: CLASSIFICATION MODELS Q1 – Q10
══════════════════════════════════════
Question 1 of 50
A 38-year-old supply chain analyst at a medical device firm is modeling the probability
that a shipped product will be returned due to defect. After fitting a logistic regression,
she sees that the coefficient for 'rush order flag' is 0.85. She needs to explain this to the
logistics director in practical terms.
A. A rush order multiplies the odds of return by approximately 2.34. ✓ CORRECT
B. The rush order flag increases the probability of return by 85 percentage points.
C. The rush order flag raises the baseline return probability from 50% to 85%.
D. The coefficient means that 85% of rush orders are returned.
Correct Answer: A
Rationale: In logistic regression, a coefficient β translates to an odds multiplier of e^β,
so e^0.85 ≈ 2.34, meaning rush orders have about 2.34 times the odds of return. Choice
B is wrong because logistic coefficients do not directly change probability by
,percentage points; the effect on probability depends on the baseline. In practice,
translating coefficients to odds ratios helps operations managers understand relative
risk without needing the full probability formula.
Question 2 of 50
A 45-year-old quality engineer at an aerospace manufacturer is building a KNN classifier
to predict component failure using 47 sensor measurements. He notices that even with
k=1, the training accuracy is only mediocre, and increasing k makes it worse.
A. He should switch to Manhattan distance because Euclidean distance fails with
continuous sensor data.
B. The model likely suffers from the curse of dimensionality, so feature selection or PCA
should be applied before KNN. ✓ CORRECT
C. Increasing k will eventually stabilize the predictions once it reaches half the sample
size.
D. The 47 sensors should be standardized by mean subtraction only, without scaling
variance.
Correct Answer: B
Rationale: KNN relies on distance calculations, and in high-dimensional spaces the
relative contrast between near and far neighbors diminishes, making feature reduction
essential. Choice C is tempting because larger k does smooth predictions, but setting k
to half the sample size would oversmooth and ignore local structure entirely. In
manufacturing, sensor arrays often produce correlated measurements, so PCA is a
common preprocessing step before distance-based classifiers.
Question 3 of 50
A 29-year-old data scientist at a renewable energy startup is training an SVM to predict
turbine failure based on two operational features: vibration frequency and oil
temperature. The scatterplot shows a clear circular decision boundary would separate
the classes best.
,A. She should use a linear kernel with a high regularization parameter to approximate
the circular boundary.
B. A polynomial kernel of degree 3 will always outperform an RBF kernel for circular
data.
C. An RBF kernel is well-suited to capture this non-linear, radial boundary without
manual feature engineering. ✓ CORRECT
D. She should add squared terms manually and then use a linear kernel to avoid
overfitting.
Correct Answer: C
Rationale: The RBF kernel implicitly maps features into an infinite-dimensional space
where complex, non-linear boundaries like circles become linearly separable. Choice D
sounds practical but requires correct specification of the transformation; if the true
boundary is circular, manually adding only squared terms might not capture the exact
geometry and adds risk of misspecification. In industrial predictive maintenance, RBF
kernels are popular because they adapt to unknown non-linear relationships between
sensor readings.
Question 4 of 50
A 52-year-old biostatistician at a hospital network is predicting readmission within 30
days using 80 patient variables. She wants a model that automatically selects the most
important predictors while shrinking others to zero.
A. She should use Ridge regularization because it handles multicollinearity among the
80 variables.
B. An unregularized logistic regression with stepwise backward selection will produce
the most stable variable selection.
C. She should use elastic net with alpha=0.5 to guarantee that all 80 variables remain in
the final model.
D. LASSO regularization is appropriate because its L1 penalty can force some
coefficients to exactly zero. ✓ CORRECT
Correct Answer: D
, Rationale: LASSO applies an L1 penalty that can shrink coefficients all the way to zero,
effectively performing variable selection alongside estimation. Choice A is incorrect
because Ridge uses an L2 penalty that shrinks coefficients but never sets them exactly
to zero, so it does not select variables. In clinical modeling with dozens of potential
predictors, LASSO helps avoid overfitting while producing interpretable, sparse models.
Question 5 of 50
A 33-year-old fraud analyst at a credit card company is building a KNN model to detect
fraudulent transactions using 'purchase amount' (range $1–$50,000) and 'merchant risk
score' (range 0–100). Her initial model performs poorly.
A. She must standardize or normalize the features so that purchase amount does not
dominate distance calculations. ✓ CORRECT
B. KNN is inherently robust to different feature scales, so the poor performance must be
due to noise.
C. The solution is to increase k until the model averages out the scale differences
between neighbors.
D. She should convert both features to binary indicators above their median values.
Correct Answer: A
Rationale: KNN uses distance metrics that are sensitive to scale, so an unscaled feature
with a large range like purchase amount will overwhelm a smaller-range feature like risk
score. Choice B is a common misconception because KNN has no built-in scale
invariance and will indeed be dominated by large-magnitude features. In fraud detection,
standardizing features ensures that subtle patterns in normalized risk scores contribute
equally to distance calculations alongside transaction amounts.
Question 6 of 50
A 41-year-old manufacturing engineer is using SVM to classify defective vs
non-defective circuit boards. He has a clean training set but suspects some labels may
Exam-Style Questions with Detailed Rationales | 100%
Verified | Pass Guaranteed – A+ Graded
TABLE OF CONTENTS
Section 1 | Classification Models | Q1 – Q10
Section 2 | Clustering and Dimensionality Reduction | Q11 – Q20
Section 3 | Time Series and Forecasting | Q21 – Q30
Section 4 | Validation and Model Selection | Q31 – Q40
Section 5 | Optimization and Linear Programming | Q41 – Q50
Instructions: Choose the single best answer. Pass: 80% in 90 minutes.
══════════════════════════════════════
SECTION 1: CLASSIFICATION MODELS Q1 – Q10
══════════════════════════════════════
Question 1 of 50
A 38-year-old supply chain analyst at a medical device firm is modeling the probability
that a shipped product will be returned due to defect. After fitting a logistic regression,
she sees that the coefficient for 'rush order flag' is 0.85. She needs to explain this to the
logistics director in practical terms.
A. A rush order multiplies the odds of return by approximately 2.34. ✓ CORRECT
B. The rush order flag increases the probability of return by 85 percentage points.
C. The rush order flag raises the baseline return probability from 50% to 85%.
D. The coefficient means that 85% of rush orders are returned.
Correct Answer: A
Rationale: In logistic regression, a coefficient β translates to an odds multiplier of e^β,
so e^0.85 ≈ 2.34, meaning rush orders have about 2.34 times the odds of return. Choice
B is wrong because logistic coefficients do not directly change probability by
,percentage points; the effect on probability depends on the baseline. In practice,
translating coefficients to odds ratios helps operations managers understand relative
risk without needing the full probability formula.
Question 2 of 50
A 45-year-old quality engineer at an aerospace manufacturer is building a KNN classifier
to predict component failure using 47 sensor measurements. He notices that even with
k=1, the training accuracy is only mediocre, and increasing k makes it worse.
A. He should switch to Manhattan distance because Euclidean distance fails with
continuous sensor data.
B. The model likely suffers from the curse of dimensionality, so feature selection or PCA
should be applied before KNN. ✓ CORRECT
C. Increasing k will eventually stabilize the predictions once it reaches half the sample
size.
D. The 47 sensors should be standardized by mean subtraction only, without scaling
variance.
Correct Answer: B
Rationale: KNN relies on distance calculations, and in high-dimensional spaces the
relative contrast between near and far neighbors diminishes, making feature reduction
essential. Choice C is tempting because larger k does smooth predictions, but setting k
to half the sample size would oversmooth and ignore local structure entirely. In
manufacturing, sensor arrays often produce correlated measurements, so PCA is a
common preprocessing step before distance-based classifiers.
Question 3 of 50
A 29-year-old data scientist at a renewable energy startup is training an SVM to predict
turbine failure based on two operational features: vibration frequency and oil
temperature. The scatterplot shows a clear circular decision boundary would separate
the classes best.
,A. She should use a linear kernel with a high regularization parameter to approximate
the circular boundary.
B. A polynomial kernel of degree 3 will always outperform an RBF kernel for circular
data.
C. An RBF kernel is well-suited to capture this non-linear, radial boundary without
manual feature engineering. ✓ CORRECT
D. She should add squared terms manually and then use a linear kernel to avoid
overfitting.
Correct Answer: C
Rationale: The RBF kernel implicitly maps features into an infinite-dimensional space
where complex, non-linear boundaries like circles become linearly separable. Choice D
sounds practical but requires correct specification of the transformation; if the true
boundary is circular, manually adding only squared terms might not capture the exact
geometry and adds risk of misspecification. In industrial predictive maintenance, RBF
kernels are popular because they adapt to unknown non-linear relationships between
sensor readings.
Question 4 of 50
A 52-year-old biostatistician at a hospital network is predicting readmission within 30
days using 80 patient variables. She wants a model that automatically selects the most
important predictors while shrinking others to zero.
A. She should use Ridge regularization because it handles multicollinearity among the
80 variables.
B. An unregularized logistic regression with stepwise backward selection will produce
the most stable variable selection.
C. She should use elastic net with alpha=0.5 to guarantee that all 80 variables remain in
the final model.
D. LASSO regularization is appropriate because its L1 penalty can force some
coefficients to exactly zero. ✓ CORRECT
Correct Answer: D
, Rationale: LASSO applies an L1 penalty that can shrink coefficients all the way to zero,
effectively performing variable selection alongside estimation. Choice A is incorrect
because Ridge uses an L2 penalty that shrinks coefficients but never sets them exactly
to zero, so it does not select variables. In clinical modeling with dozens of potential
predictors, LASSO helps avoid overfitting while producing interpretable, sparse models.
Question 5 of 50
A 33-year-old fraud analyst at a credit card company is building a KNN model to detect
fraudulent transactions using 'purchase amount' (range $1–$50,000) and 'merchant risk
score' (range 0–100). Her initial model performs poorly.
A. She must standardize or normalize the features so that purchase amount does not
dominate distance calculations. ✓ CORRECT
B. KNN is inherently robust to different feature scales, so the poor performance must be
due to noise.
C. The solution is to increase k until the model averages out the scale differences
between neighbors.
D. She should convert both features to binary indicators above their median values.
Correct Answer: A
Rationale: KNN uses distance metrics that are sensitive to scale, so an unscaled feature
with a large range like purchase amount will overwhelm a smaller-range feature like risk
score. Choice B is a common misconception because KNN has no built-in scale
invariance and will indeed be dominated by large-magnitude features. In fraud detection,
standardizing features ensures that subtle patterns in normalized risk scores contribute
equally to distance calculations alongside transaction amounts.
Question 6 of 50
A 41-year-old manufacturing engineer is using SVM to classify defective vs
non-defective circuit boards. He has a clean training set but suspects some labels may
