CS 559 Quiz 9: Combining Models Questions
and Answers | 2026 Update | 100% Correct.
1. What is the primary goal of combining multiple models?
A) To increase training time
B) To reduce model interpretability
C) To improve predictive performance and robustness
D) To eliminate the need for cross-validation
Answer: C
Rationale: Combining models leverages diverse strengths, reduces bias/variance, and
often yields better generalization than any single model.
2. Which technique averages predictions from several models trained on different
subsets of data?
A) Boosting
B) Bagging
C) Stacking
D) Bayesian model averaging
Answer: B
Rationale: Bagging (Bootstrap Aggregating) trains base learners on bootstrapped
samples and averages their outputs.
,3. In boosting, how are successive models trained?
A) Independently on random data subsets
B) Sequentially, with each focusing on previous errors
C) In parallel on the full dataset
D) By averaging parameters of earlier models
Answer: B
Rationale: Boosting assigns higher weights to misclassified instances, so subsequent
models correct prior mistakes.
4. Which combination method uses a meta-learner to learn optimal weights for
base predictions?
A) Bagging
B) Random forest
C) Stacking
D) AdaBoost
Answer: C
Rationale: Stacking trains a secondary model on the outputs of base models to produce
the final prediction.
5. Bayesian model averaging (BMA) combines models by:
A) Averaging predictions with equal weights
B) Weighting each model by its posterior probability
,C) Choosing the model with the highest likelihood
D) Using cross-validation to select the best model
Answer: B
Rationale: BMA weights models according to their posterior probabilities given the data.
6. Which of the following is NOT an ensemble method?
A) Random forest
B) Gradient boosting
C) Support vector machine (SVM)
D) AdaBoost
Answer: C
Rationale: SVM is a single classifier; random forest, gradient boosting, and AdaBoost are
ensembles.
7. What is the main advantage of combining models over selecting a single best
model?
A) Simpler implementation
B) Lower computational cost
C) Reduced overfitting and improved stability
D) Guaranteed optimal performance
Answer: C
Rationale: Ensembles smooth out individual errors and often reduce variance without
increasing bias excessively.
, 8. In a mixture of experts, what determines which expert is used for a given input?
A) A gating network
B) The average of all experts
C) Random selection
D) The expert with the largest training set
Answer: A
Rationale: A gating network learns to assign weights to experts based on the input
features.
9. Which ensemble method is most susceptible to overfitting if base learners are
too complex?
A) Bagging
B) Random forest
C) Boosting
D) Averaging
Answer: C
Rationale: Boosting iteratively focuses on hard examples, which can overfit if the base
learners are strong.
10. Random forest reduces overfitting by:
A) Using deep decision trees
B) Averaging many trees built on bootstrapped samples with random feature selection
and Answers | 2026 Update | 100% Correct.
1. What is the primary goal of combining multiple models?
A) To increase training time
B) To reduce model interpretability
C) To improve predictive performance and robustness
D) To eliminate the need for cross-validation
Answer: C
Rationale: Combining models leverages diverse strengths, reduces bias/variance, and
often yields better generalization than any single model.
2. Which technique averages predictions from several models trained on different
subsets of data?
A) Boosting
B) Bagging
C) Stacking
D) Bayesian model averaging
Answer: B
Rationale: Bagging (Bootstrap Aggregating) trains base learners on bootstrapped
samples and averages their outputs.
,3. In boosting, how are successive models trained?
A) Independently on random data subsets
B) Sequentially, with each focusing on previous errors
C) In parallel on the full dataset
D) By averaging parameters of earlier models
Answer: B
Rationale: Boosting assigns higher weights to misclassified instances, so subsequent
models correct prior mistakes.
4. Which combination method uses a meta-learner to learn optimal weights for
base predictions?
A) Bagging
B) Random forest
C) Stacking
D) AdaBoost
Answer: C
Rationale: Stacking trains a secondary model on the outputs of base models to produce
the final prediction.
5. Bayesian model averaging (BMA) combines models by:
A) Averaging predictions with equal weights
B) Weighting each model by its posterior probability
,C) Choosing the model with the highest likelihood
D) Using cross-validation to select the best model
Answer: B
Rationale: BMA weights models according to their posterior probabilities given the data.
6. Which of the following is NOT an ensemble method?
A) Random forest
B) Gradient boosting
C) Support vector machine (SVM)
D) AdaBoost
Answer: C
Rationale: SVM is a single classifier; random forest, gradient boosting, and AdaBoost are
ensembles.
7. What is the main advantage of combining models over selecting a single best
model?
A) Simpler implementation
B) Lower computational cost
C) Reduced overfitting and improved stability
D) Guaranteed optimal performance
Answer: C
Rationale: Ensembles smooth out individual errors and often reduce variance without
increasing bias excessively.
, 8. In a mixture of experts, what determines which expert is used for a given input?
A) A gating network
B) The average of all experts
C) Random selection
D) The expert with the largest training set
Answer: A
Rationale: A gating network learns to assign weights to experts based on the input
features.
9. Which ensemble method is most susceptible to overfitting if base learners are
too complex?
A) Bagging
B) Random forest
C) Boosting
D) Averaging
Answer: C
Rationale: Boosting iteratively focuses on hard examples, which can overfit if the base
learners are strong.
10. Random forest reduces overfitting by:
A) Using deep decision trees
B) Averaging many trees built on bootstrapped samples with random feature selection
