All subjects · Information Technology · WIT13 2 hour written exam on data, enabling technologies and systems development

Machine learning: supervised, unsupervised and its uses

Edexcel IAL Information Technology · WIT13 2 hour written exam on data, enabling technologies and systems development

Machine learning lets computers improve at a task by learning from data rather than following only hand-written rules. You need supervised and unsupervised learning and the impact of natural language processing, speech, image and pattern recognition (spec 16.1).

Practise this in the app: instant marking and a fix for every mistake. Free, no email needed to start.

Practise this lesson

What machine learning is

Machine learning (ML) is a branch of artificial intelligence in which a system learns patterns from data and uses them to make predictions or decisions about new data, improving as it is given more data.

Typical process: collect and clean data; split it into a training set and a test set; train a model on the training data; test it on unseen data to measure accuracy; deploy it; monitor and retrain as data changes.

Features and functions: finds complex patterns humans would miss; adapts to new data; automates decisions at scale. Needs large amounts of good-quality, representative data and significant computing power.

Supervised and unsupervised learning

Supervised learningUnsupervised learning
DataA **labelled** data set: each example has the correct answer (photo labelled 'cat'; email labelled 'spam')An **unlabelled** (unknown) data set: no correct answers given
GoalLearn the link between inputs and labels to predict labels for new dataDiscover hidden structure: groups, patterns or unusual items
TasksClassification (spam or not), regression (predict a house price)Clustering (customer segments), association, anomaly detection
ExamplesImage recognition, spam filters, credit scoring, medical diagnosisGrouping customers by behaviour, detecting unusual network traffic

Uses and impact

Impact: automation of routine work, new services, better accessibility (captions, voice control), faster diagnosis. Concerns: bias learned from unrepresentative data, lack of transparency (hard to explain decisions), privacy (training on personal data), job losses in some roles, errors made with confidence, accountability.

Worked example

A spam filter is tested on 100 emails: 45 spam correctly flagged, 40 genuine correctly passed, 5 genuine wrongly flagged, 10 spam missed. Find its accuracy.

  1. Correct = 45 + 40 = 85
  2. Accuracy = 85 / 100 = 0.85 = 85%
  3. Note: 5 genuine emails were wrongly flagged (false positives), which could mean important mail is missed

Answer: Accuracy 85%

Key idea: ML learns patterns from data. Supervised: labelled data, predicts labels (classification, regression). Unsupervised: unlabelled data, finds groups and anomalies. Uses: NLP, speech, image and pattern recognition. Concerns: bias, transparency, privacy, jobs.

Check yourself

  1. A model is trained on 10 000 X-ray images, each labelled 'fracture' or 'no fracture'. Which type of learning is this?
    • Unsupervised learning
    • Supervised learning
    • Expert system rules
    • Batch processing
    Show answer

    Supervised learning. The training data is labelled.

  2. A retailer's system groups customers by buying behaviour without being told what the groups should be. Which type of learning is this?
    • Supervised learning
    • Rule-based programming
    • Transaction processing
    • Unsupervised learning
    Show answer

    Unsupervised learning. No labels; groups are discovered.

  3. A system reads thousands of product reviews and classifies each as positive or negative. Which ML application is this?
    • Image recognition
    • Natural language processing
    • Speech recognition
    • Fleet management
    Show answer

    Natural language processing. Understanding written language.

  4. How does a supervised image recognition system learn to recognise dogs?
    • It stores one photo of a dog and checks for an exact match
    • A programmer writes a rule for every possible dog
    • It is trained on many photographs labelled 'dog' and 'not dog' and learns the features that distinguish them
    • It asks the user each time
    Show answer

    It is trained on many photographs labelled 'dog' and 'not dog' and learns the features that distinguish them. Training on labelled examples builds a general model.

  5. A model classifies 200 images and gets 170 right. What is its accuracy as a percentage?
    Show answer

    85 %. 170 / 200 x 100 = 85%.

  6. A recruitment model trained mostly on past successful applicants from one group rejects other groups more often. What is the issue?
    • A hardware fault
    • Too much documentation
    • Bias learned from unrepresentative training data
    • A network partition
    Show answer

    Bias learned from unrepresentative training data. Models reproduce patterns in their data.

  7. Why is a machine learning model tested on data it was not trained on?
    • To check that it works on new, unseen examples rather than just memorising the training data
    • To make training faster
    • To reduce storage
    • To label the training data
    Show answer

    To check that it works on new, unseen examples rather than just memorising the training data. Testing on unseen data measures real performance.

  8. Explain one benefit of speech recognition for individuals.
    Show answer

    It lets people control devices or write text by speaking, so they can use technology hands-free while driving or if they have a disability that makes typing difficult.. Speaking instead of typing, so hands-free or accessible.

Practise with instant feedback

Sources

Last updated 8 November 2026. Spotted a mistake? Tell us.