S5 ICT: AI and Machine Learning, evaluation and training set-up
- 2026-11-20
S5 ICT: AI and Machine Learning, evaluation and training set-up
This paper works through data splits, accuracy, precision and recall, then tests judgement on learning rate, overfitting, cross validation and missing values.
Intermediate: metrics and training set-up Numbers drive this paper: an 80 to 20 split of 500 gives 400 and 100, ninety correct out of one hundred is accuracy 0.9, forty true positives with ten false positives gives precision 0.8, five false negatives give recall 0.889, a 37 point gap between training and test signals overfitting, 200 samples over eight features is 25 each, ten fold cross validation tests 100 at a time, and five per cent of 2 000 records is 100. It also covers instability from a large learning rate, deeper trees trading bias for variance, and the split between regression and classification.
01 Paper details (check answers afterwards)
| Item | Detail |
|---|---|
| Questions | 10 |
| Suggested time | 25 |
| Difficulty mix | Foundation 3 · Intermediate 3 · Challenge 3 · Error redo 1 |
| Chapter | Chapter: S5 ICT: Artificial Intelligence and Machine Learning |
| Level | Intermediate (10 items) |
02 Questions (10 items)
- FoundationIf 500 samples are split 80 to 20, how many go to training and testing?
- FoundationIf 90 of 100 test samples are classified correctly, what is the accuracy?
- FoundationWith 40 true positives, 10 false positives, 5 false negatives and 45 true negatives, what are precision and recall?
- IntermediateWhat happens when the learning rate is set too high?
- IntermediateTraining accuracy is 99 per cent but test accuracy only 62 per cent. What is the problem?
- IntermediateWith eight features and 200 samples, how many samples are there per feature?
- ChallengeWhat does a deeper decision tree do to bias and variance?
- ChallengeWith 1 000 samples under ten fold cross validation, how large is each fold and each training set?
- ChallengeWhich task type fits predicting flat prices, and which fits detecting spam mail?
- Error redoHow many of 2 000 records are affected when five per cent of values are missing?
03 Answers and one-line notes
| # | Answer | One-line note |
|---|---|---|
| 1 | 400 for training and 100 for testing. | 400 samples is eighty per cent, and the remaining 100 are held back to test generalisation. |
| 2 | The accuracy is 0.90, that is ninety per cent. | Accuracy is the number of correct predictions divided by the number of test samples. |
| 3 | Precision 0.80 and recall 0.889. | Precision is 40 over 50 and recall is 40 over 45, dividing by predicted positives and actual positives respectively. |
| 4 | Each update moves too far, so the loss oscillates and may fail to converge. | The parameters keep overshooting the minimum on both sides, so training never settles. |
| 5 | It is overfitting: the model memorised the training samples and does not generalise. | The gap of 37 percentage points is a clear sign of a generalisation failure. |
| 6 | Twenty five. | 200 divided by 8 is 25, and few samples per feature makes overfitting likely. |
| 7 | Bias falls and variance rises. | A deeper tree fits the training data more closely but its predictions swing more on new data. |
| 8 | Each fold has 100 samples for testing and 900 for training. | 1 000 divided by 10 is 100, and the other nine folds give 900 training samples. |
| 9 | Predicting prices is regression and spam detection is classification. | Regression outputs a continuous value while classification outputs a discrete class. |
| 10 | One hundred records. | 2 000 times 0.05 is 100, and those values need filling or dropping before training. |
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.
Want to compute precision and recall by hand? Book a TutorZone tutor and work through all ten items in 25 minutes.Tutor 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.



