PE1-6.4 Machine Learning: Supervised & Unsupervised
You can apply linear and polynomial regression, KNN, decision trees, naive Bayes and support vector machines, and use K-means clustering, PCA and boosting and bagging ensembles.
The supervised methods are best learned by what boundary each draws: KNN draws whatever the data says locally, a decision tree draws axis-aligned boxes, an SVM draws the widest possible margin. Naive Bayes keeps working despite an independence assumption that is almost always false, and knowing why that is is a genuinely useful piece of statistical intuition. On the unsupervised side, PCA and K-means both reduce description length — one in dimensions, one in points — and both need you to choose a number the data does not supply.
Work through these
Introduction to machine learning; types of learning and applications
The frame for the whole topic: what learning from data means and the three broad kinds. Getting the kinds straight tells you what data you need before choosing a method.
Linear regression and polynomial regression
Fitting a line, and then a curve. It is the simplest supervised method and the one where overfitting is easiest to see and understand.
KNN classifier
Classifying by looking at what is nearby, with no training step at all. Its cost is at prediction time, which is the opposite of everything else here.
Decision trees
Splitting the data by one feature at a time, producing a model you can read. Being explainable is its main advantage and it is a real one.
Naive Bayes classifier
Classifying with probabilities and a strong independence assumption. The assumption is usually false and the method works anyway, which is worth understanding.
Support vector machines
Separating classes with the widest possible margin, and the trick that handles curved boundaries. It was the strongest method available before neural networks returned.
K-means clustering
The first unsupervised method: grouping without labels. Choosing the number of groups is the honest difficulty, and it has no clean answer.
Dimensionality reduction: principal component analysis
Reducing the number of features while keeping the variation. It is used constantly for visualisation and for making other methods tractable.
Ensemble learning: boosting and bagging
Combining many weak models into a strong one, in two different ways. It is what wins most competitions on data of this kind, so it earns its place.
Sign in to keep your progress.
Free resources
We haven't checked most of these for screen reader use yet.
Links last checked 30 Aug 2026.
Stuck here?
Ask a mentor. A real person answers, and they can see exactly which topic you're on. Usually within a couple of working days.
Checking your session…
Topics shown in module order.