Learning from labelled data to predict labels (classification) or values (regression).
Unsupervised learning
Finding patterns in unlabelled data, e.g. clustering.
Reinforcement learning
An agent learns by trial and error through rewards and penalties.
Overfitting
A model learns the training data too closely, including noise, and performs poorly on new data.
Training / test data
Data used to fit the model / unseen data used to evaluate it.
Workflow: collect data → clean and preprocess (missing values, normalization, encoding) → split into training and test sets → train → evaluate → deploy and monitor.
The full notes, worked example, interactive diagram and check questions for this subtopic are in the app.
Exam tip: Always say which type of learning a scenario uses and why (labels or no labels, rewards).
A4.2Machine learning ethics
Key terms
Algorithmic bias
Systematic unfair outcomes caused by biased data or design.
Transparency
How understandable a model's decisions are (black-box vs explainable AI).
Accountability
Who is responsible when an automated decision causes harm.
Bias in training data (e.g. under-representation of some groups) leads to biased predictions in hiring, lending, policing or facial recognition.
The full notes, worked example and check questions for this subtopic are in the app.
Exam tip: Structure ethics answers: issue → who is affected → consequence → mitigation.