01
True or false
1 point
True or false: Compared with squared-error metrics, mean absolute error is less affected by large prediction errors.
Free Practice Quiz Question List
7 Model Evaluation Online Quiz Questions
Use this free practice quiz with 20 questions to review 7 Model Evaluation, test your knowledge, and prepare for your next test or exam.
20 questions
02
Choose one
1 point
A team has separate training, validation, and test sets. What is the intended role of the test set?
- A
To tune model settings repeatedly during development
- B
To provide a final estimate of performance after model selection
- C
To fit preprocessing separately for every candidate model
- D
To determine which metric should be used after inspecting test results
03
Written response
1 point
In six-fold cross-validation, how many folds are used to train the model in each round? Enter the number of folds.
04
Fill in the blank
1 point
When unusually large prediction errors are especially costly, may be more informative than MAE.
05
Choose one
1 point
A screening system must minimize missed positive cases, even if it sends some extra follow-up alerts. Which metric should the team emphasize?
- A
Accuracy
- B
Precision
- C
Recall
- D
ROC AUC
06
Choose all
1 point
Which quantities are counts recorded in a binary confusion matrix? Select all that apply.
- A
True positives
- B
False positives
- C
True negatives
- D
False negatives
- E
Accuracy
- F
F1 score
07
Written response
1 point
What is the name of the classification metric defined as the fraction of predictions that are correct?
08
Fill in the blank
1 point
If a dataset contains repeated observations from the same people or other units, keep the separated across folds.
09
True or false
1 point
True or false: A negative R² means a regression model performs worse than predicting the test-set mean.
- A
True
- B
False
10
Choose all
1 point
For multiclass precision, recall, or F1, which statements correctly describe averaging methods? Select all that apply.
- A
Macro averaging gives each class equal weight.
- B
Weighted averaging weights classes by their frequency.
- C
Macro averaging weights classes by their frequency.
- D
Weighted averaging gives each class equal weight regardless of frequency.
11
Choose one
1 point
A classifier outputs scores rather than only class labels. What does ROC AUC summarize?
- A
The model’s performance at one fixed threshold chosen using the test set
- B
The model’s ranking performance across decision thresholds
- C
The fraction of predictions that are correct at every threshold
- D
The average recall across classes
12
Open ended
1 point
Why can overall accuracy give a misleading impression on an imbalanced classification task? Name a metric that gives each class equal weight and explain what it adds.
13
Choose one
1 point
A team needs a simple baseline for a classification model. Which prediction rule is given as an example?
- A
Predict a different randomly selected class for every test example
- B
Predict the most frequent class in the training set
- C
Predict the class that gives the highest test-set accuracy after repeated tuning
- D
Predict the least frequent class in the validation set
14
Choose one
1 point
When comparing two candidate models, which evaluation practice is recommended?
- A
Give each model a different test set so each can be evaluated under its strongest conditions
- B
Use the same data splits for each model
- C
Choose a different metric for each model after viewing test results
- D
Use test-set results repeatedly to choose model settings
15
Choose one
1 point
A regression system occasionally makes very large errors, and those errors are especially costly. Which metric is more informative for emphasizing this risk?
- A
MAE, because it gives large errors more influence than other errors.
- B
RMSE, because it penalizes large errors more heavily than MAE.
- C
R², because a negative value means large errors are costly.
- D
Accuracy, because it reports error in the target's units.
16
Written response
1 point
What classification metric is the fraction of examples predicted positive that are truly positive?
17
Choose one
1 point
A team needs to tune model settings and then report a final performance estimate. Which use of the validation and test sets best limits an overly optimistic estimate?
- A
Use the test set to tune settings, then use the validation set for the final unbiased estimate.
- B
Use the training set for the final estimate and the test set to select models.
- C
Use the validation set to select models or tune settings, and reserve the test set for final evaluation.
- D
Use the test set repeatedly for model selection so its estimate becomes more reliable.
18
Choose one
1 point
A classifier is evaluated on a dataset where one class is much more common than the others. Why might its accuracy give a misleading impression of performance?
- A
A model can achieve high accuracy by predicting the majority class while missing most minority-class cases.
- B
Accuracy gives every class equal weight, regardless of its frequency.
- C
High accuracy guarantees that the model finds most examples from every class.
- D
Accuracy measures only the fraction of predicted positive examples that are truly positive.
19
True or false
1 point
ROC AUC summarizes ranking performance across decision thresholds, but it does not determine which threshold is best for a particular application.
- A
True
- B
False
20
Written response
1 point
In multiclass evaluation, which averaging method weights each class equally rather than weighting classes by their frequency?