Skip to content

Topic Drill 3.3.1: Calculating ML Functional Performance Metrics (K3) ​

About this drill

LO AI-3.3.1 (K3): Calculate common ML functional performance metrics from a given set of confusion matrix data.
Lesson: Ch 3, section 3.3.1

This drill covers every point in section 3.3 / 3.3.1:

  • The confusion matrix: TP, TN, FP, FN, and both layouts (predicted rows / actual rows)
  • Accuracy, precision, recall (sensitivity) and F1-score: formulas and calculation
  • Interpretation: "how sure about positives" vs. "how confident no positives are missed"; F1 as the harmonic mean; what a low F1 means
  • Picking the right metric for a context, and why accuracy misleads on imbalanced data (interpretation is supplementary)

Work it on paper

Exam questions give the matrix as a table. Here the matrices are written out in brackets, so sketch each one on paper first, label TP/FP/FN/TN, then calculate. That is exactly the routine you need under exam conditions.

Topic Drill: 3.3.1 (K3)

Q1. A disease-screening model was tested on 200 patients. [Matrix A, rows = ACTUAL, columns = PREDICTED: Actual Positive → 36 predicted positive, 4 predicted negative; Actual Negative → 12 predicted positive, 148 predicted negative] What is its PRECISION?

A90.0%
B92.0%
C75.0%
D25.0%

Q2. Same screening model. [Matrix A, rows = ACTUAL, columns = PREDICTED: Actual Positive → 36 predicted positive, 4 predicted negative; Actual Negative → 12 predicted positive, 148 predicted negative] What is its RECALL?

A90.0%
B75.0%
C10.0%
D92.5%

Q3. Same screening model. [Matrix A, rows = ACTUAL, columns = PREDICTED: Actual Positive → 36 predicted positive, 4 predicted negative; Actual Negative → 12 predicted positive, 148 predicted negative] What is its ACCURACY?

A75.0%
B90.0%
C81.8%
D92.0%

Q4. Same screening model: precision = 75%, recall = 90%. What is its F1-score (rounded)?

A82.5
B81.8
C92.0
D67.5

Q5. A defect prediction model flags files as "defective" (positive). [Matrix B, rows = PREDICTED, columns = ACTUAL: Predicted Positive → 30 actually positive, 20 actually negative; Predicted Negative → 10 actually positive, 140 actually negative] Which option represents the CORRECT formula for RECALL?

A(30 / 50) × 100
B(170 / 200) × 100
C(30 / 40) × 100
D(20 / 50) × 100

Q6. Same defect predictor. [Matrix B, rows = PREDICTED, columns = ACTUAL: Predicted Positive → 30 actually positive, 20 actually negative; Predicted Negative → 10 actually positive, 140 actually negative] Which option represents the CORRECT formula for PRECISION?

A(30 / 50) × 100
B(30 / 40) × 100
C(140 / 160) × 100
D(10 / 40) × 100

Q7. For a cancer-screening model, missing a patient who has cancer is far more harmful than sending a healthy patient for a follow-up check. Which metric should the acceptance criteria emphasize MOST?

AAccuracy
BPrecision
CSpecificity
DRecall

Q8. For a spam filter, putting a genuine customer email into the spam folder is considered much worse than letting some spam through. Which metric matters MOST?

ARecall
BPrecision
CAccuracy
DNeuron coverage

Q9. Of 500 transactions, 5 are fraudulent (positive). A model predicts "not fraud" for every transaction. What are its accuracy and recall?

AAccuracy 99%, recall 0%
BAccuracy 1%, recall 100%
CAccuracy 99%, recall 100%
DAccuracy 0%, recall 99%

Q10. Which TWO statements about the F1-score are CORRECT according to the syllabus?

Select 2 options.

AF1 always equals the arithmetic average of precision and recall
BF1 is the harmonic mean of precision and recall
CF1 uses the number of true negatives
DA low F1 indicates the model struggles to identify positives, either missing true cases or generating many false alarms
EF1 is always higher than both precision and recall

Q11. Model X has precision 90% and recall 60%. Model Y has precision 75% and recall 75%. Which model has the higher F1-score?

AModel X, because its precision is higher
BNeither: both F1-scores are 75, because the averages are equal
CModel Y: F1 = 75 for Y vs. 72 for X
DModel X: F1 = 75 for X vs. 72 for Y

0 / 11 answered

← 3.2.3 Datasets · Next drill: 3.4.1 Deep Neural Networks →

Unofficial study guide for the ISTQB® Certified Tester AI Testing (CT-AI) exam. Not affiliated with International Software Testing Qualifications Board (ISTQB®).