Precision, recall and F1 quiz

Question 1

In binary classification, what does 'precision' measure?

Accepted Answer

The fraction of predicted positives that are actually positive

Answer

The fraction of actual positives the model correctly identified

Answer

The fraction of all predictions (positive and negative) that are correct

Answer

The fraction of actual negatives correctly identified by the model

Question 2

In binary classification, what does 'recall' (also called sensitivity) measure?

Accepted Answer

The fraction of actual positives that the model correctly identified

Answer

The fraction of predicted positives that are truly positive

Answer

The overall accuracy of all predictions made by the model

Answer

The fraction of negative predictions that turn out to be correct

Question 3

The F1 score is defined as which type of mean of precision and recall?

Accepted Answer

Harmonic mean

Answer

Arithmetic mean

Answer

Geometric mean

Answer

Weighted sum

Question 4

In a binary confusion matrix, a 'False Positive' (FP) occurs when:

Accepted Answer

The model predicts positive but the true label is negative

Answer

The model correctly predicts a positive case (TP)

Answer

The model predicts negative but the true label is positive (FN)

Answer

The model predicts negative and the true label is also negative (TN)

Question 5

Why is the harmonic mean used in the F1 score instead of the arithmetic mean?

Accepted Answer

The harmonic mean strongly penalises extreme imbalances between precision and recall

Answer

The harmonic mean is always greater than or equal to the arithmetic mean

Answer

The harmonic mean is computationally cheaper than the arithmetic mean

Answer

The harmonic mean equals precision whenever recall is exactly 1.0

Question 6

In which application should recall be prioritised over precision?

Accepted Answer

Cancer screening, where missing a real cancer case is life-threatening

Answer

Spam email filtering, where marking a legitimate email as spam is very costly

Answer

Movie recommendation, where suggesting a bad film is merely annoying

Answer

Ad targeting, where showing an irrelevant ad wastes advertiser budget

Question 7

For a spam email filter, why is precision generally more important than recall?

Accepted Answer

Flagging a legitimate email as spam (FP) means the user may never see it

Answer

Missing a spam email (FN) causes the user to permanently lose important data

Answer

Spam emails always contain malware so recall is automatically perfect

Answer

Recall is always higher than precision in all modern spam filters

Question 8

A model has precision = 0.8 and recall = 0.8. What is its F1 score?

Accepted Answer

0.80

Answer

0.64

Answer

0.72

Answer

0.88

Question 9

If a model achieves precision = 1.0 but recall = 0.0, what is its F1 score?

Accepted Answer

0.0

Answer

0.5

Answer

1.0

Answer

Undefined — division by zero

Question 10

In the Fβ score formula Fβ = (1+β²)·P·R / (β²·P + R), what does setting β > 1 mean?

Accepted Answer

Recall is valued more than precision

Answer

Precision is valued more than recall

Answer

Both metrics are weighted equally regardless of the value of β

Answer

The score becomes undefined when β exceeds 1

Question 11

How is macro-averaged F1 computed in multi-class classification?

Accepted Answer

F1 is computed per class, then averaged with equal weight for each class

Answer

All TP, FP, FN are pooled across classes and a single F1 is computed from those totals

Answer

F1 is computed per class, then averaged weighted by the number of samples per class

Answer

The F1 of the largest class is reported as the macro F1

Question 12

For micro-averaged F1 in multi-class classification, what is done first?

Accepted Answer

All TP, FP, FN are summed across every class and then one F1 is computed from these totals

Answer

F1 is computed per class and then averaged

Answer

Only the majority class is used to compute F1

Answer

Accuracy is computed first and used as a stand-in for micro F1

Question 13

In a 3-class problem, the per-class F1 scores are 0.90, 0.20, and 0.10. What is the macro-averaged F1?

Accepted Answer

0.40

Answer

0.90

Answer

0.60

Answer

0.50

Question 14

A dataset has 950 negatives and 50 positives. A model predicts 'negative' for every single input. Which statement is TRUE?

Accepted Answer

Accuracy = 95% and F1 for the positive class = 0, so accuracy is misleading here

Answer

Both accuracy and F1 equal 95%, confirming the model is genuinely good

Answer

F1 = 95% and accuracy = 0%, so F1 is the misleading metric here

Answer

Neither metric can be computed without additional confusion matrix details

Question 15

A cancer-detection model correctly identifies 70 out of 100 actual cancer patients. What is its recall?

Accepted Answer

0.70

Answer

0.30

Answer

0.50

Answer

Cannot be determined without knowing the number of false positives

Question 16

A classifier predicts 50 positives, of which 40 are correct. There are 100 actual positives in the dataset. What are precision and recall?

Accepted Answer

Precision = 0.8, Recall = 0.4

Answer

Precision = 0.4, Recall = 0.8

Answer

Precision = 0.8, Recall = 0.8

Answer

Precision = 0.5, Recall = 0.5

Question 17

Which sklearn function prints precision, recall, F1, and support for every class in a single call?

Accepted Answer

sklearn.metrics.classification_report

Answer

sklearn.metrics.accuracy_score

Answer

sklearn.metrics.confusion_matrix

Answer

sklearn.metrics.roc_auc_score

Question 18

When you lower the decision threshold from 0.50 to 0.20 (predict positive when model confidence > 20%), what typically happens to precision and recall?

Accepted Answer

Recall increases; precision decreases

Answer

Recall decreases; precision increases

Answer

Both recall and precision increase together

Answer

Both recall and precision decrease together

Question 19

For a perfect binary classifier that makes zero mistakes, what are precision and recall?

Accepted Answer

Precision = 1.0, Recall = 1.0

Answer

Precision = 1.0, Recall = 0.0

Answer

Precision = 0.5, Recall = 0.5

Answer

Precision = 0.0, Recall = 1.0

Question 20

A model predicts 'diabetic' for every patient. The true prevalence of diabetes in the dataset is 10%. What are recall and precision?

Accepted Answer

Recall = 1.0, Precision = 0.1

Answer

Recall = 0.1, Precision = 1.0

Answer

Recall = 1.0, Precision = 1.0

Answer

Recall = 0.1, Precision = 0.1

Question 21

In medical testing, 'sensitivity' and 'recall' are used interchangeably. Which formula defines both?

Accepted Answer

TP / (TP + FN)

Answer

TP / (TP + FP)

Answer

TN / (TN + FP)

Answer

TN / (TN + FN)

Question 22

Which formula computes F1 directly from TP, FP, and FN without first computing precision and recall separately?

Accepted Answer

2·TP / (2·TP + FP + FN)

Answer

TP / (TP + FP + FN)

Answer

TP / (2·TP + FP + FN)

Answer

2·TP / (TP + FP + FN)

Question 23

A dataset has Class A with 10 samples and Class B with 990 samples. How do macro F1 and weighted F1 differ?

Accepted Answer

Macro F1 treats both classes equally; weighted F1 gives Class B roughly 99× more influence

Answer

Weighted F1 treats both classes equally; macro F1 weights by class size

Answer

Both are identical for any class size distribution

Answer

Macro F1 is always higher than weighted F1 for imbalanced datasets

Question 24

Under what condition does the F1 score equal both precision and recall exactly?

Accepted Answer

When precision and recall happen to be equal to each other

Answer

When the dataset is perfectly balanced between the two classes

Answer

When the model predicts every sample as positive

Answer

When the model achieves 100% overall accuracy

Question 25

A model has TP = 50, FP = 10, and FN = 20. Which value is closest to its F1 score?

Accepted Answer

0.769

Answer

0.625

Answer

0.714

Answer

0.833

Question 26

Which special property of the harmonic mean makes it uniquely appropriate for combining precision and recall into F1?

Accepted Answer

It equals zero if either precision or recall is zero, correctly scoring useless models at zero

Answer

It is always at least as large as the arithmetic mean, rewarding good models

Answer

It always equals the geometric mean when both values lie between 0 and 1

Answer

It weights precision twice as much as recall by default

Question 27

In a spam filter, a False Negative (FN) means:

Accepted Answer

A spam email passes the filter and lands in the user's inbox

Answer

A legitimate email is flagged as spam and disappears from the inbox (FP)

Answer

The filter correctly identifies and quarantines a spam message (TP)

Answer

The filter correctly passes a legitimate message to the inbox (TN)

Question 28

A model has F1 = 0.75 and precision = 0.90. What is the recall? (Solve algebraically.)

Accepted Answer

0.643

Answer

0.50

Answer

0.60

Answer

0.825

Question 29

A classifier is set with a very high decision threshold (e.g., 0.97). What is the typical effect on precision and recall?

Accepted Answer

Very high precision, very low recall

Answer

Very high recall, very low precision

Answer

Both precision and recall converge toward 0.5

Answer

F1 increases because both metrics improve simultaneously

Question 30

In sklearn's f1_score(), which value of the 'average' parameter treats every class as equally important regardless of how many samples it has?

Accepted Answer

average='macro'

Answer

average='micro'

Answer

average='weighted'

Answer

average='binary'

Test yourself