A4.3 Machine learning approaches: notes and practice

What you need to know

Our summary of the topic in our own words, not the IB guide. Points marked HL are for Higher Level only.

• Explain how linear regression fits a line to data to predict a continuous valueHL
• Interpret a fitted regression line and use it to make a predictionHL
• Describe supervised classification techniques and what kind of problem each suitsHL
• Evaluate a classifier using accuracy and the counts of correct and incorrect predictions per classHL
• Explain what a hyperparameter is and how it differs from a learned parameterHL
• Explain how hyperparameter tuning affects underfitting and overfittingHL
• Explain what clustering does and how it differs from classificationHL
• Describe how a clustering algorithm assigns points to groups and how the number of groups is chosenHL
• Explain association rule learning and interpret support and confidence for a ruleHL
• Apply association rule thinking to a basket-style data setHL
• Describe how a reinforcement learning agent learns from states, actions and rewardsHL
• Explain the exploration and exploitation trade-offHL
• Describe how a genetic algorithm uses a population, a fitness function, selection, crossover and mutationHL
• Apply a genetic algorithm design to a described optimisation problemHL
• Describe the structure of an artificial neural network: input, hidden and output layers, weights and activationHL
• Explain at a conceptual level how a network learns by adjusting weights from errorHL
• Describe the parts of a convolutional neural network and why convolution suits image dataHL
• Explain the role of pooling and of successive layers in building up image featuresHL
• Compare candidate models for a task and select one, justifying the choiceHL
• Explain why a model must be evaluated on data it was not trained onHL

Practice questions

Original questions written for the 2027 course, marked as you go. Easiest first.

• Using a linear regression line for salesWritten · Stretch
• Accuracy and errors of a phishing filterWritten · Stretch
• Decision tree depth and overfittingWritten · Stretch
• Testing a bike-share demand model fairlyWritten · Stretch

Exam tips for A4.3

• Know what problem each method suits. Many questions describe a task and ask which approach fits.
• Explain overfitting simply: the model learns the training data too well and does worse on new data.
• Use the words layer, weight and activation correctly when describing a neural network.