Back to Subject

Unit 6: MODEL PERFORMANCE - Practice Quiz

INT234 — Predictive Analytics 60 Questions
0 Correct 0 Wrong 60 Left
0/60

1 What does bias describe in a predictive model?

Bias-variance trade-off Easy
A. Error from random data changes
B. Number of features in the model
C. Error from overly simple assumptions
D. Size of the training dataset

2 What does variance describe in a predictive model?

Bias-variance trade-off Easy
A. Dependence on a fixed model assumption
B. Sensitivity to changes in training data
C. Number of classes in the target
D. Count of errors in the test set

3 A model with high bias is usually described as what?

Bias-variance trade-off Easy
A. Overfitting the training data
B. Using too many training sets
C. Averaging many independent models
D. Underfitting the training data

4 A model with high variance is usually described as what?

Bias-variance trade-off Easy
A. Overfitting the training data
B. Using a constant prediction
C. Ignoring the target variable
D. Underfitting the training data

5 What is the main goal of managing the bias-variance trade-off?

Bias-variance trade-off Easy
A. To remove the target variable
B. To increase the training error
C. To improve performance on new data
D. To reduce the number of observations

6 Which model behavior is most likely to have low bias and high variance?

Bias-variance trade-off Easy
A. A simple linear regression model
B. A constant-value prediction model
C. A very complex decision tree
D. A model with no input variables

7 What usually happens to bias when model complexity increases?

Bias-variance trade-off Easy
A. Bias generally decreases
B. Bias generally increases
C. Bias always becomes zero
D. Bias becomes unrelated to complexity

8 What is ensemble learning?

Ensemble Learning: Basic Concept of Bagging and Boosting Easy
A. Training one model without validation
B. Removing variables from a dataset
C. Using only the largest training sample
D. Combining several models for prediction

9 What is the main idea of bagging?

Ensemble Learning: Basic Concept of Bagging and Boosting Easy
A. Remove randomness from model training
B. Train one model with increasing depth
C. Train models on different data samples
D. Use only incorrectly classified cases

10 How are predictions commonly combined in bagging for classification?

Ensemble Learning: Basic Concept of Bagging and Boosting Easy
A. By using the highest feature value
B. By selecting the first model
C. By majority voting
D. By removing all minority classes

11 How are predictions commonly combined in bagging for regression?

Ensemble Learning: Basic Concept of Bagging and Boosting Easy
A. By counting the predicted classes
B. By averaging predictions
C. By choosing the largest prediction
D. By choosing the smallest prediction

12 What is the main idea of boosting?

Ensemble Learning: Basic Concept of Bagging and Boosting Easy
A. Replace all missing values with class labels
B. Build models independently from identical data
C. Use only a single model for every prediction
D. Build models sequentially to correct errors

13 In boosting, what receives more attention during later training stages?

Ensemble Learning: Basic Concept of Bagging and Boosting Easy
A. Observations predicted incorrectly
B. Observations with missing feature names
C. Observations selected completely at random
D. Observations with the largest identifiers

14 Which method trains its base models independently and in parallel?

Ensemble Learning: Basic Concept of Bagging and Boosting Easy
A. Normalization
B. Bagging
C. Pruning
D. Boosting

15 What is a base learner in an ensemble?

Ensemble Learning: Basic Concept of Bagging and Boosting Easy
A. The final test dataset
B. A performance measurement
C. An individual model in the ensemble
D. A missing value indicator

16 What type of models are combined to form a random forest?

Random forests Easy
A. Multiple database tables
B. Multiple decision trees
C. Multiple linear equations
D. Multiple target variables

17 A random forest primarily uses which ensemble method?

Random forests Easy
A. Clustering
B. Scaling
C. Bagging
D. Boosting

18 What additional source of randomness is commonly used in a random forest?

Random forests Easy
A. Random subsets of features
B. Random output labels
C. Random target values
D. Random evaluation metrics

19 How does a random forest usually make a classification prediction?

Random forests Easy
A. By majority vote of its trees
B. By averaging the class names alphabetically
C. By using the deepest tree only
D. By selecting the first tree's result

20 How does a random forest usually make a regression prediction?

Random forests Easy
A. By averaging its tree predictions
B. By counting the number of features
C. By selecting the first observed value
D. By choosing its smallest tree

21 A model has very low training error but substantially higher validation error. Which diagnosis is most appropriate?

Bias-variance trade-off Medium
A. Low bias and low variance
B. High bias and high variance
C. High bias and low variance
D. Low bias and high variance

22 A data scientist reduces the maximum depth of a decision tree. What is the most likely effect?

Bias-variance trade-off Medium
A. Both bias and variance decrease
B. Bias increases and variance decreases
C. Both bias and variance increase
D. Bias decreases and variance increases

23 Two models have the following validation results. Model A has training MSE and validation MSE . Model B has training MSE and validation MSE . Which model is more likely to generalize better?

Bias-variance trade-off Medium
A. Model B, because validation error is less important
B. Model A, because its validation error is lower
C. Model A, because it has higher model complexity
D. Model B, because its training error is lower

24 A highly regularized linear model underfits both the training and validation data. Which change would most directly address this problem?

Bias-variance trade-off Medium
A. Reduce the training set
B. Increase the validation error
C. Increase model flexibility
D. Add more regularization

25 For squared-error prediction, the expected test error can be decomposed into bias, variance, and irreducible noise. Which action primarily reduces variance?

Bias-variance trade-off Medium
A. Increase systematic model assumptions
B. Use a larger and representative training set
C. Remove informative predictors from the model
D. Add measurement noise to the response variable

26 A model's validation error is high because it is too simple to capture nonlinear relationships. Which pattern should be expected when model complexity is gradually increased?

Bias-variance trade-off Medium
A. Training error rises while validation error stays constant
B. Training error rises while validation error always rises
C. Training error stays constant while validation error always falls
D. Training error falls while validation error may first fall then rise

27 What is the main principle behind bagging?

Ensemble Learning: Basic Concept of Bagging and Boosting Medium
A. Train models on bootstrap samples and aggregate predictions
B. Train one model using all transformed features
C. Train models sequentially on previous errors
D. Train models only on incorrectly classified cases

28 A bagging ensemble contains 100 regression trees. How are their predictions usually combined?

Ensemble Learning: Basic Concept of Bagging and Boosting Medium
A. The average of the tree predictions is calculated
B. The largest tree prediction is selected
C. The prediction with the highest training error is selected
D. The prediction from the deepest tree is selected

29 In boosting, a new weak learner is generally trained to address which issue?

Ensemble Learning: Basic Concept of Bagging and Boosting Medium
A. Features removed from the original data
B. Random samples excluded from every bootstrap set
C. Predictions made by an unrelated test set
D. Errors made by the current ensemble

30 A classification ensemble combines the outputs of several classifiers by majority vote. If 7 classifiers predict class A and 3 predict class B, what is the ensemble prediction?

Ensemble Learning: Basic Concept of Bagging and Boosting Medium
A. Both classes, because the vote is uncertain
B. Class A, based on the majority vote
C. Class B, based on the minority vote
D. No class, because the classifiers disagree

31 Why can bagging reduce the variance of unstable models such as decision trees?

Ensemble Learning: Basic Concept of Bagging and Boosting Medium
A. It removes all features from the training data
B. It averages predictions from differently sampled models
C. It increases the bias of every base learner
D. It forces every tree to have identical errors

32 Which situation is more suitable for boosting than for standard bagging?

Ensemble Learning: Basic Concept of Bagging and Boosting Medium
A. The goal is to correct residual errors sequentially
B. The model should ignore difficult observations
C. The base learners must be trained independently
D. The main concern is reducing variance in deep trees

33 A boosting model performs extremely well on training data but poorly on validation data after many iterations. Which adjustment is most reasonable?

Ensemble Learning: Basic Concept of Bagging and Boosting Medium
A. Make every weak learner fully identical
B. Remove the validation data from evaluation
C. Reduce iterations or apply stronger regularization
D. Increase the number of boosting iterations

34 What is a key operational difference between bagging and boosting?

Ensemble Learning: Basic Concept of Bagging and Boosting Medium
A. Bagging increases errors, while boosting removes labels
B. Bagging uses test data, while boosting uses training data
C. Bagging always uses linear models, while boosting uses trees
D. Bagging is usually parallel, while boosting is sequential

35 A random forest differs from ordinary bagging of decision trees mainly because it also:

Random forests Medium
A. Trains every tree on the same observations
B. Uses a random subset of features at each split
C. Uses only one tree for final prediction
D. Prevents trees from using bootstrap samples

36 Why does reducing correlation among trees generally improve a random forest?

Random forests Medium
A. Averaging less-correlated errors lowers ensemble variance
B. Averaging less-correlated trees eliminates irreducible noise
C. Less-correlated trees require no training data
D. Correlated trees always increase classification bias

37 A random forest classifier is trained using bootstrap samples. What are out-of-bag observations useful for?

Random forests Medium
A. Replacing the target variable during training
B. Estimating performance without a separate validation set
C. Guaranteeing that every tree has zero training error
D. Increasing the number of features at every split

38 A random forest has high training accuracy but low validation accuracy. Which parameter change is most likely to help reduce overfitting?

Random forests Medium
A. Allow unlimited tree depth
B. Increase the minimum samples required at a leaf
C. Use fewer observations for validation
D. Increase the number of candidate features without testing

39 In a random forest regression model, 5 trees predict , , , , and . What is the forest prediction using simple averaging?

Random forests Medium
A.
B.
C.
D.

40 A dataset contains 50 predictors, but only a few are informative. What is one benefit of selecting a random subset of predictors at each random forest split?

Random forests Medium
A. It forces every tree to use the same split variables
B. It removes the need to tune any forest parameters
C. It guarantees that every predictor is informative
D. It makes trees more diverse and limits dominant predictors

41 A model's expected test error is decomposed as . A change reduces squared bias by 12 units, increases variance by 7 units, and leaves irreducible noise unchanged. What is the net change in expected test error?

Bias-variance trade-off Hard
A. It decreases by 19 units
B. It increases by 19 units
C. It decreases by 5 units
D. It increases by 5 units

42 Two models have the following error components: Model A has bias squared and variance ; Model B has bias squared and variance . If both face the same irreducible noise, which model has lower expected test error?

Bias-variance trade-off Hard
A. Model B, because its total reducible error is lower
B. Model A, because its variance is lower
C. Model A, because its total reducible error is lower
D. Model B, because its bias squared is lower

43 A highly flexible predictor achieves nearly zero training error, but its test error changes substantially across different training samples. Which diagnosis is most consistent with this behavior?

Bias-variance trade-off Hard
A. High bias and high variance
B. Low bias and low variance
C. Low bias and high variance
D. High bias and low variance

44 Suppose the data-generating process contains substantial irreducible noise. A more complex model reduces systematic prediction error but produces only a negligible reduction in total test error. Which explanation is most appropriate?

Bias-variance trade-off Hard
A. The noise term dominates the remaining error
B. The training error must exceed the test error
C. The model's bias must increase with complexity
D. The model has eliminated the sampling variance

45 A regularization parameter is increased in a regression model. Which pattern is generally expected, assuming all other conditions remain fixed?

Bias-variance trade-off Hard
A. Training bias decreases and variance increases
B. Both training bias and variance increase
C. Training bias increases and variance decreases
D. Both training bias and variance decrease

46 A learning curve shows training error continuing to fall as the sample size grows, while validation error remains nearly constant and substantially higher. Which intervention is most likely to address the dominant problem?

Bias-variance trade-off Hard
A. Collect more identical labels
B. Increase regularization strength
C. Increase model flexibility
D. Reduce the feature representation

47 Why does averaging predictions from many independently trained high-variance regressors often improve generalization?

Ensemble Learning: Basic Concept of Bagging and Boosting Hard
A. Averaging removes all individual model bias
B. Averaging cancels uncorrelated prediction fluctuations
C. Averaging increases each model's training accuracy
D. Averaging converts regression into classification

48 For base regressors with identical variance and pairwise error correlation , which condition makes increasing provide the greatest variance reduction?

Ensemble Learning: Basic Concept of Bagging and Boosting Hard
A. A larger base-model bias
B. Large positive
C. Small positive
D. Zero or negative

49 A bagged classifier is built from unstable decision trees. If every bootstrap sample produces nearly identical trees, what is the most likely outcome?

Ensemble Learning: Basic Concept of Bagging and Boosting Hard
A. A large reduction in ensemble variance
B. A limited reduction in ensemble variance
C. Guaranteed elimination of overfitting
D. A large increase in ensemble bias

50 In boosting, a new weak learner is fitted primarily to examples that the current ensemble predicts poorly. What is the central purpose of this procedure?

Ensemble Learning: Basic Concept of Bagging and Boosting Hard
A. To decorrelate bootstrap samples
B. To maximize the variance of weak learners
C. To reduce the ensemble's residual error
D. To discard all previously fitted learners

51 A boosting model's training loss decreases monotonically, but its validation loss begins increasing after many iterations. Which response is most directly justified?

Ensemble Learning: Basic Concept of Bagging and Boosting Hard
A. Increase the number of boosting iterations
B. Replace sequential fitting with bootstrap sampling
C. Increase the learning rate without validation
D. Use early stopping or stronger shrinkage

52 Which comparison correctly distinguishes standard bagging from standard boosting?

Ensemble Learning: Basic Concept of Bagging and Boosting Hard
A. Bagging requires weak learners; boosting requires deep trees
B. Bagging reweights errors; boosting averages bootstrap models
C. Bagging trains independently; boosting updates later learners
D. Bagging is sequential; boosting is parallel

53 A dataset contains mislabeled observations that repeatedly receive large weights during boosting. What risk is most likely to increase?

Ensemble Learning: Basic Concept of Bagging and Boosting Hard
A. The ensemble overemphasizes label noise
B. The ensemble becomes insensitive to outliers
C. The training loss becomes irreducible noise
D. The base learners become statistically independent

54 Why does a random forest randomly restrict the candidate features considered at each split?

Random forests Hard
A. To reduce correlation among trees in the ensemble
B. To reduce tree diversity and speed convergence
C. To increase correlation among tree errors
D. To ensure every tree uses every feature

55 A random forest uses fully grown, unpruned trees and a large number of trees. Which statement best describes the expected behavior?

Random forests Hard
A. The forest cannot estimate generalization error internally
B. The forest has high bias because trees are independently trained
C. The forest must have high variance because each tree is deep
D. The forest can generalize well despite low-bias base trees

56 In a random forest, an observation is out-of-bag for a particular tree when:

Random forests Hard
A. Its features are excluded at every tree split
B. It is selected more than once by that tree
C. Its response value is missing during training
D. It is excluded from that tree's bootstrap sample

57 Why can out-of-bag error provide a useful estimate of random-forest generalization error?

Random forests Hard
A. Each observation is evaluated using the full training forest
B. Each observation is removed from all model-fitting procedures
C. Each observation is evaluated by trees that excluded it
D. Each observation is evaluated only by trees that trained on it

58 A random forest shows high training accuracy and high out-of-bag error. Increasing the number of trees has little effect. Which explanation is most plausible?

Random forests Hard
A. The forest has insufficient averaging
B. The bootstrap samples are necessarily identical
C. The feature-label relationship may be poorly learnable
D. The forest's trees are too weak to fit patterns

59 A feature has high random-forest impurity importance because it is selected early in many trees. The feature is strongly correlated with a second feature. Which conclusion is safest?

Random forests Hard
A. The feature must be measured without error
B. The feature is uniquely causal
C. The feature has no predictive information
D. The importance may be shared or unstable

60 A random forest is trained on a dataset where one feature has thousands of possible split points while another has only two. What bias may affect impurity-based feature importance?

Random forests Hard
A. Preference for the feature with more split points
B. Preference determined only by feature names
C. Preference for the feature with fewer split points
D. Equal importance regardless of split opportunities
Previous MCQ
Unit 5