Model evaluation metrics quiz

Question 1

What is accuracy in classification, and how is it calculated?

Accepted Answer

(TP + TN) / (TP + TN + FP + FN)

Answer

TP / (TP + FP)

Answer

TP / (TP + FN)

Answer

(FP + FN) / total

Question 2

What is a confusion matrix and what do its four cells represent for binary classification?

Accepted Answer

A table of TP, TN, FP, FN comparing predicted vs actual labels

Answer

A plot of accuracy over training epochs

Answer

A matrix of feature correlations

Answer

A table of model hyperparameters

Question 3

Define precision and give its formula.

Accepted Answer

TP / (TP + FP)

Answer

TP / (TP + FN)

Answer

(TP + TN) / total

Answer

TN / (TN + FP)

Question 4

Define recall (sensitivity) and give its formula.

Accepted Answer

TP / (TP + FN)

Answer

TP / (TP + FP)

Answer

TN / (TN + FP)

Answer

FP / (FP + TN)

Question 5

What is the F1 score and why is it useful?

Accepted Answer

2 * (precision * recall) / (precision + recall)

Answer

(precision + recall) / 2

Answer

precision * recall

Answer

precision + recall

Question 6

Why can accuracy be misleading for an imbalanced dataset?

Accepted Answer

A trivial majority-class predictor can score high accuracy while ignoring the rare class

Answer

Accuracy cannot be computed when classes are imbalanced

Answer

Accuracy always equals recall on imbalanced data

Answer

Imbalance makes accuracy exceed 100%

Question 7

What is specificity and how is it calculated?

Accepted Answer

TN / (TN + FP)

Answer

TP / (TP + FN)

Answer

TP / (TP + FP)

Answer

TN / (TN + FN)

Question 8

In medical screening for a serious disease, would you prioritize precision or recall, and why?

Accepted Answer

Recall, because missing a true case (false negative) is very costly

Answer

Precision, because false alarms must be eliminated entirely

Answer

Accuracy, because it always balances both

Answer

Specificity, because negatives matter most

Question 9

In an email spam filter, why might you prioritize precision over recall?

Accepted Answer

Precision, to avoid wrongly flagging real emails as spam

Answer

Recall, to catch every possible spam message

Answer

Accuracy, since both errors are equal

Answer

Specificity, since spam is the negative class

Question 10

What does an ROC curve plot?

Accepted Answer

True Positive Rate vs False Positive Rate across thresholds

Answer

Precision vs recall across thresholds

Answer

Accuracy vs epochs

Answer

Loss vs learning rate

Question 11

What does AUC (Area Under the ROC Curve) measure, and what value indicates random guessing?

Accepted Answer

AUC = 0.5 means random guessing; 1.0 is perfect

Answer

AUC = 0 means random guessing; 1.0 is perfect

Answer

AUC = 1.0 means random guessing; 0.5 is perfect

Answer

AUC = 0.5 means perfect classification

Question 12

What is a Type I error in the context of classification?

Accepted Answer

A false positive (predicting positive when actually negative)

Answer

A false negative (predicting negative when actually positive)

Answer

A correct positive prediction

Answer

A correct negative prediction

Question 13

What is a Type II error in the context of classification?

Accepted Answer

A false negative (predicting negative when actually positive)

Answer

A false positive (predicting positive when actually negative)

Answer

A correct negative prediction

Answer

A correct positive prediction

Question 14

What does Mean Absolute Error (MAE) measure in regression?

Accepted Answer

The average of absolute errors |y_pred - y_true|

Answer

The average of squared errors

Answer

The square root of average squared errors

Answer

The proportion of correct predictions

Question 15

What is Mean Squared Error (MSE) and why does it penalize large errors heavily?

Accepted Answer

Average of squared errors; squaring inflates large errors

Answer

Average of absolute errors; treats all errors equally

Answer

Proportion of correct predictions

Answer

Average of the predictions themselves

Question 16

How does RMSE relate to MSE, and what advantage does it give?

Accepted Answer

RMSE = sqrt(MSE), giving an error in the target's units

Answer

RMSE = MSE squared

Answer

RMSE = MSE / n

Answer

RMSE = MSE - MAE

Question 17

What does the R-squared (coefficient of determination) metric tell you?

Accepted Answer

The fraction of target variance explained by the model

Answer

The average absolute prediction error

Answer

The number of correct classifications

Answer

The model's training time

Question 18

What is log loss (cross-entropy) and what kind of predictions does it require?

Accepted Answer

A loss on predicted probabilities that penalizes confident wrong answers

Answer

A metric that requires only hard 0/1 labels

Answer

The square root of MSE

Answer

The fraction of correctly ranked pairs

Question 19

A model has TP=40, FP=10, FN=20, TN=30. What is its precision?

Accepted Answer

0.8

Answer

0.67

Answer

0.57

Answer

0.4

Question 20

A model has TP=40, FP=10, FN=20, TN=30. What is its recall?

Accepted Answer

0.67

Answer

0.8

Answer

0.57

Answer

0.5

Question 21

What is the difference between macro-averaging and micro-averaging a metric across classes?

Accepted Answer

Macro averages per-class scores equally; micro pools counts and favors large classes

Answer

Macro pools counts; micro averages per class equally

Answer

They always give identical results

Answer

Macro only works for regression

Question 22

How does changing the classification threshold affect precision and recall?

Accepted Answer

A higher threshold usually raises precision but lowers recall

Answer

A higher threshold raises both precision and recall

Answer

The threshold has no effect on either metric

Answer

A higher threshold always lowers precision

Question 23

What is balanced accuracy and when is it preferred over plain accuracy?

Accepted Answer

The mean of per-class recall; better for imbalanced data

Answer

The total correct over total samples

Answer

The square root of accuracy

Answer

Accuracy computed only on the majority class

Question 24

Why should evaluation metrics be measured on a test set rather than the training set?

Accepted Answer

The test set estimates generalization to unseen data, avoiding overfitting bias

Answer

Training metrics are always lower than test metrics

Answer

The test set is used to train the final model

Answer

Metrics can only be computed on training data

Question 25

How can comparing training and test metrics help detect overfitting?

Accepted Answer

A large train-minus-test performance gap signals overfitting

Answer

Equal train and test scores always mean overfitting

Answer

Overfitting makes test scores higher than train scores

Answer

Overfitting cannot be detected from metrics

Question 26

What is k-fold cross-validation and why is it used for evaluation?

Accepted Answer

Splitting data into k folds, testing on each once, and averaging scores

Answer

Training k separate models on the full dataset

Answer

Using k features to evaluate the model

Answer

Running the model k times on the same test set

Question 27

Why is a baseline (such as a dummy classifier predicting the majority class) useful when evaluating a model?

Accepted Answer

It sets a reference score the model must beat to be considered useful

Answer

It guarantees the model will not overfit

Answer

It replaces the need for a test set

Answer

It is the same as the F1 score

Question 28

What is the F-beta score and how does beta change its behavior?

Accepted Answer

Beta > 1 weights recall more; beta < 1 weights precision more

Answer

Beta always weights precision more

Answer

Beta only changes the scale, not the balance

Answer

Beta = 1 ignores recall entirely

Question 29

What does MAPE (Mean Absolute Percentage Error) measure and what is one limitation?

Accepted Answer

Average percentage error; it breaks down when actual values are near zero

Answer

Average squared error; it is undefined for negative values

Answer

Proportion of correct predictions; it ignores magnitude

Answer

The square root of MSE; it has no limitations

Question 30

How does the precision-recall curve differ from the ROC curve, and when is it more informative?

Accepted Answer

PR plots precision vs recall and is better for imbalanced positive classes

Answer

PR plots TPR vs FPR like ROC but rescaled

Answer

ROC is always better for imbalanced data

Answer

PR and ROC give identical information

Test yourself