中文（繁體）
中文（简体）
English
S6 ICT: Artificial Intelligence and Machine Learning – Practice Paper 3 (Challenge, 10 Questions)
S6 ICT: Artificial Intelligence and Machine Learning – Practice Paper 3 (Challenge, 10 Questions)
Paper three tackles three-way splits, ten-fold cross validation, choosing between two thresholds with unequal error costs, majority voting across five models reaching 94.2%, data leakage and halving the learning rate.
Quick answer Challenge items weigh costs and process: compare total cost first, then check whether preprocessing ever touched the test set.
01 Paper details (check answers afterwards)
| Item | Detail |
|---|---|
| Questions | 10 questions |
| Suggested time | 30 minutes |
| Difficulty mix | Foundation 1 · Intermediate 2 · Challenge 3 · Error redo 4 |
| Chapter | Chapter: S6 ICT | Artificial Intelligence and Machine Learning |
| Level | Challenge |
02 Questions (10 items)
- Foundation1 000 images are split 70/15/15 into training, validation and test sets. How many images in each?
- IntermediateWith ten-fold cross validation on 600 images, each fold holds 60 images for testing. How large is the training set per fold, and how many runs in total?
- IntermediateA false alarm costs 2 dollars and a miss costs 50 dollars. Threshold A gives 14 false positives and 4 false negatives; threshold B gives 30 and 2. How much cheaper is B?
- ChallengeFive models each reach 80.0% accuracy and their errors are independent. Using majority voting, what is the accuracy?
- ChallengeModel X gives 45 true positives, 5 false positives and 15 false negatives. What are its precision and recall?
- ChallengeIf the cost of a miss is far higher than the cost of a false alarm, which metric should be raised first?
- Error redoA student computes the mean and standard deviation on the whole data set before splitting, then standardises. What problem follows?
- Error redoA student raises training epochs from 20 to 200; test accuracy falls from 65.0% to 58.0%. He says the model simply needs more training. Judge this.
- Error redoAfter changing the class counts from 900 to 100 into 500 each, the same model drops from 90.0% to 78.0% accuracy, and the student says the model got worse. Is this right?
- Error redoThe learning rate is 0.01 and is halved every 10 epochs. What is it after 30 epochs?
03 Answers and one-line notes
| # | Answer | One-line note |
|---|---|---|
| 1 | 700, 150 and 150. | 1 000 times 0.7 is 700, and 1 000 times 0.15 is 150. |
| 2 | 540 images per fold, ten runs in total. | 600 minus 60 is 540, and the ten folds take turns as the test set. |
| 3 | B is 68 dollars cheaper. | A costs 14 times 2 plus 4 times 50, which is 228, while B costs 30 times 2 plus 2 times 50, which is 160. |
| 4 | About 94.2%. | The chance of three or more correct votes is about 0.942, so voting repairs many single-model errors. |
| 5 | Precision 90.0% and recall 75.0%. | 45 divided by 50 is 0.90, and 45 divided by 60 is 0.75. |
| 6 | Recall. | A miss is a false negative, and raising recall cuts misses at the price of more false alarms. |
| 7 | Data leakage, so the test score looks too optimistic. | Test-set statistics have already shaped preprocessing, which is the same as seeing the answers early. |
| 8 | Wrong, this is overfitting. | Longer training fits the training set ever more closely while test performance drops. |
| 9 | Not necessarily, the old figure was inflated by imbalance. | Always predicting the majority class already gives 90.0% at 900 to 100, so precision and recall should be compared. |
| 10 | 0.00125. | Halving 0.01 three times gives 0.005, then 0.0025 and finally 0.00125. |
04 What to do after marking
- After marking, redo only the wrong items and write down which step went wrong (formula, unit, or misread number).
- Repeat the same item two days later; move to the next paper only when both attempts are correct.
- Parents only need the answer and note columns; there is no need to explain every item.
Find a tutor freeTutor Registration →
Questions and answers are written originally by the TutorZone team and are not taken from any school, publisher or examination authority paper. Every value is checked by computer.