Wronged Questions: Statistical Learning

Question 1

Adding more predictions will never (increase/decrease) R^2

Answer 1

Decrease

Question 2

T/F: The expected test MSE formula applies to both quantitative and qualitative responses

Answer 2

False. It only applies to quantitative responses

Question 3

T/F: In the classification setting, the bias-variance trade-off does not apply since y_i is quantitative.

Answer 3

False. The bias-variance trade-off does indeed apply in the classification setting, albeit with some modifications due to
being categorical rather than quantitative.

Question 4

T/F: The training error rate is defined as the proportion of correct classifications made when applying our estimate
to the training observations.

Answer 4

False. The training error rate is defined as the proportion of incorrect classifications made when applying our estimate
to the training observations.

Question 5

T/F: A classifier’s effectiveness is determined by the magnitude of its training error rate rather than its test error rate.

Answer 5

False. A classifier’s effectiveness is determined by the magnitude of its test error rate rather than its training error rate.

Question 6

T/F: The Bayes classifier is known to produce the highest possible test error rate, known as the Bayes error rate.

Answer 6

False. The Bayes classifier is known to produce the lowest possible test error rate, known as the Bayes error rate.

Question 7

T/F: The Bayes error rate serves a role similar to that of the irreducible error in the classification setting.

Answer 7

True. The Bayes error rate is analogous to the irreducible error in classification, representing the lowest error rate that can be achieved by any classifier and is due to the noise in the data itself.

Question 8

Less flexible, more interpretable

Answer 8

Lasso, subset selection

Question 9

Moderately flexible and interpretable

Answer 9

Least squares, regression trees, classification trees

Question 10

More flexible, less interpretable

Answer 10

Bagging, boosting

Question 11

T/F: The K-Nearest Neighbor algorithm determines the function for f
that includes free parameters.

Answer 11

False. KNN is non-parametric

Question 12

T/F: Random forest makes no assumption about f’s function form.

Answer 12

True, random forest is non-parametric

Question 13

T/F: Compared to non-parametric methods, parametric methods are more versatile in fitting various forms of f.

Answer 13

False. Non-parametric methods are more versatile because they make no assumptions about f’s form.

Question 14

T/F: Compared to non-parametric methods, parametric methods typically need a larger number of observations to accurately estimate f.

Answer 14

False. Parametric methods need smaller sample sizes because they already assume a pattern

Question 15

T/F: Compared to non-parametric methods, parametric methods are generally more difficult to interpret.

Answer 15

False, parametric methods are easier to interpret due to them already having built-in structure

Question 16

T/F: Decision boundaries become overly flexible and capture noise, especially at lower K values.

Answer 16

True

Question 17

T/F: Larger K values lead to smoother, less flexible boundaries that better generalize to new data.

Answer 17

True

Question 18

T/F: Smaller K values result in boundaries that closely follow the training data, potentially capturing noise.

Answer 18

True

Question 19

T/F: Larger K values result in highly complex boundaries that can easily overfit the data.

Answer 19

False. Smaller K values result in highly complex boundaries that can easily overfit the data.

Question 20

T/F: Smaller K values create smoother and more generalized boundaries.

Answer 20

False. Larger K values create smoother and more generalized boundaries.

Question 21

Inference

Answer 21

Exploring + confirming the significant associations between predictors and the response

Question 22

Non-parametric learning methods

Answer 22

1) KNN
2) Decision trees
3) Bagging, boosting, random forest

Question 23

Bias

Answer 23

Error arising from the assumptions made in the statistical learning tool

Question 24

Variance

Answer 24

Error arising from the model’s sensitivity towards the training data

Question 25

T/F: If we are mainly interested in inference, then restrictive models are much more interpretable.

Answer 25

True

Question 26

T/F: Subset selection is more interpretable than linear regression.

Answer 26

True

Question 27

T/F: Bagging and boosting are more restrictive than linear regression.

Answer 27

False

Question 28

T/F: Highly flexible methods make it difficult to discern how changes in predictors affect the response.

Answer 28

True

Question 29

T/F: The reducible error provides a lower bound for the test error.

Answer 29

False

Question 30

T/F: In the classification setting, the bias-variance trade-off does not apply since yi is quantitative.

Answer 30

False. The bias-variance trade-off does indeed apply in the classification setting, albeit with some modifications due to yi being categorical rather than quantitative.

Question 31

T/F: KNN can often produce classifiers that are close to the optimal Bayes classifier.

Answer 31

True