Decision Trees and Tree Ensemble Methods Quiz

Auto audio on: the next questions will be read aloud when you click Continue.

1. 1

   What is the Gini impurity of a decision-tree node in which every observation belongs to the same class?

   A 0 B 0.5 C 1
2. 2

   What is the usual effect of substantially increasing a decision tree's maximum depth?

   A It increases training bias and decreases variance B It decreases training bias but increases variance C It guarantees better performance on unseen data
3. 3

   How is information gain for a candidate decision-tree split calculated?

   A Weighted child entropy minus parent entropy B Parent entropy plus weighted child entropy C Parent entropy minus weighted child entropy
4. 4

   In the illustrated tree, which leaf is reached by a sample with x1 = 4 and x2 = 3?

   

   A Leaf L2 B Leaf L1 C Leaf L3
5. 5

   What is the primary statistical benefit of bagging many high-variance decision trees?

   A It eliminates all bias from each tree B It reduces variance by averaging predictions C It converts every nonlinear model into a linear model
6. 6

   The diagram shows a random forest workflow. Which added mechanism distinguishes a random forest from ordinary bagging of decision trees?

   

   A Trees are trained sequentially on previous errors B Every tree must have a depth of exactly one C Each split considers a random subset of features
7. 7

   Why does selecting a random subset of features at each random-forest split often improve the ensemble?

   A It reduces correlation among the trees B It forces every tree to produce identical predictions C It prevents the trees from using informative features
8. 8

   The illustrated bootstrap sample was drawn from records 1 through 5. Which record is out-of-bag for this tree?

   

   A Record 1 B Record 3 C Record 5
9. 9

   How does boosting differ fundamentally from bagging?

   A Boosting trains independent models only on disjoint feature sets B Boosting always averages fully grown trees with equal weights C Boosting trains learners sequentially to address earlier errors
10. 10

    In gradient boosting for regression, what is the next tree generally trained to approximate?

    

    A The current model's negative loss gradients, often represented as pseudo-residuals B A random permutation of the target values C Only the predictions already produced by the first tree
11. 11

    What is a common consequence of decreasing the learning rate in gradient-boosted trees while tuning the model appropriately?

    A Fewer trees are always required B More trees are usually required, but generalization may improve C All new trees are prevented from affecting predictions
12. 12

    After a training observation is misclassified by a weak learner in AdaBoost, what usually happens to its weight?

    A It is permanently set to zero B It remains fixed while only feature weights change C It increases relative to correctly classified observations
13. 13

    The three panels show decision boundaries fitted to the same noisy training data. Which panel most strongly suggests overfitting?

    

    A Panel C B Panel A C Panel B
14. 14

    According to the permutation-importance chart, which feature is most important to the fitted model?

    

    A Feature A B Feature B C Feature C
15. 15

    What does early stopping typically determine when training a gradient-boosted tree model?

    A The alphabetical order of the input features B The bootstrap sample used by every future tree C The number of boosting iterations based on validation performance