Explain machine learning
Describe how examples can change a program's process or data and why results need evaluation.
Section 6 · Automated and emerging technologies
Exam scope and assessment notes
This sequence follows Cambridge IGCSE Computer Science 0984, 2026–2028 syllabus Version 4. It covers subject content sections 6.1–6.3 only. The separately published 2029–2031 syllabus is not mapped here. Practice is original, self-checked and does not award an official Cambridge grade. Full timed papers and human subject-teacher review remain outstanding. Cambridge states that 0984 has the same 2026–2028 subject content as 0478, with 9–1 grading. In June and November, Cambridge lists 0984 in Administrative zone 3 and for centres in Oman and the UAE. Confirm availability for your centre and exam series with Cambridge International. The separately published 2029–2031 syllabus changes to Python 3 only and removes pseudocode; this 2026–2028 route does not cover those changes.
Before you begin
- No programming language is required. Be comfortable with place value and multiplication; each lesson introduces the computer-science notation it uses.
Work on paper. Try each question before revealing help, and explain why your method works.
Quick readiness check
What can a machine-learning program adapt?
What you will learn
- Define machine learning in syllabus terms
- Explain adaptation from data
- Describe the need for representative testing
Machine learning adapts a program
Machine learning is when a program can automatically adapt its own processes and/or data. It uses examples or feedback to adjust how it responds to future inputs.
Apply and check the rule
For example, a classifier can be trained on labelled messages and update the pattern it uses to sort new messages. The result depends on the examples: missing, unbalanced or inaccurate data can lead to poor or unfair classifications. Test with representative examples and review important decisions.
Pause and explain
Why test with new examples after training?
Worked example
A spam filter learns from messages labelled by users. What changes as it adapts?
Show the worked solution
- The filter receives examples with labels.
- It adjusts its process or data to recognise patterns.
- It applies the updated pattern to new messages.
Answer The filter adjusts its process or data using labelled examples and applies the result to later messages.
Try a topic test
Choose Easy, Normal or Difficult. Save your answers and open worked review after finishing.
Test this topic →From guided practice to a new situation
Write your answer and reasoning first. Use a hint only if you are stuck. The model solution is for self-checking; your written work is not automatically marked.
Describe the steps by which a photo classifier could learn to separate cats from dogs.
Give me a hint
Use labelled examples and a changed classification process.
Compare my reasoning
- Provide images labelled cat or dog.
- The program adjusts its process using patterns in the examples.
- Test the adapted model on new images.
Train with labelled images, adapt the classification process, then evaluate it on images not used for training.
Look for these in your work
- Example data
- Adaptation
- Unseen evaluation
A classifier was trained mostly on daytime images. Explain one risk when it is used on night-time images.
Give me a hint
Compare the training data with the new conditions.
Compare my reasoning
- Night images may differ from the examples.
- The learned pattern may not generalise to different lighting.
- Collect representative data and test performance in night conditions.
The model may perform poorly because its examples did not represent night conditions. Test and train with appropriate night images.
Look for these in your work
- Training gap
- Effect
- Improvement
A company says that an adaptive model cannot be biased because it learns automatically. Evaluate this claim.
Give me a hint
Automatic adaptation does not make data neutral.
Compare my reasoning
- The model learns from selected examples and feedback.
- Incomplete or biased data can produce biased patterns.
- Audit representative cases and maintain human review for consequential decisions.
The claim is false: learning reflects the data and design choices. Audit results on varied representative examples and retain human review where errors could harm people.
Look for these in your work
- Learning source
- Bias risk
- Responsible evaluation
Common mistakes
- Saying machine learning needs no data
- Assuming more examples always make a system fair or accurate
What can you explain now?
What is central to machine learning in the syllabus?
Compare with the explanation
A program automatically adapts its processes or data
Learning changes how a program uses information or operates.
After trying it yourself, choose your next review. This is your self-assessment.
Your review choice appears on Today. Sign in to sync it across devices.
Make it stick
What changes when a machine-learning program adapts, and why should it be tested?
If you needed the solution, close it and solve the problem again from a blank page. A correct answer is stronger when you can explain the reason for each step.