03 · Machine-learning algorithms
Learn what each model is for, how it works and when it can mislead. Start with a familiar prediction problem, then compare approaches using sound evaluation.
Choose by the problem you want to solve
Thirteen learning paths, including Linear Regression and evaluation. The order below is a guide, not a route lock.
A house-price question is the easiest place to see how a model learns from data.
Explain a numerical prediction and how its settings are learned.
9 lessons and activities · Equations, errors, updates, multiple inputs, implementation and limits. Recommended first algorithm.
Open Linear RegressionDo this next. Every later model is compared using the same ideas: what the model may know, how data is split and what a score means.
Define what a model can legitimately know and what its score means.
6 lessons · Prediction time, split boundaries, errors, probability quality and uncertainty.
Open Evaluation FoundationsAdd curved features, then see how coefficient penalties keep a flexible model honest.
Explain how additional input features can represent a curve.
4 lessons · Curved features, flexibility and validation. Helpful foundation: Linear Regression.
Open Polynomial RegressionExplain how coefficient penalties change a fitted model.
6 lessons · Regularization, scaling, sparsity and careful selection.
Open RegularizationCompare different ways to reason about a named outcome, without treating predictions as causal guarantees.
Connect a score to probability and a separate decision.
6 lessons · Log loss, thresholds, regularization, workflows and limits.
Open Logistic RegressionExplain a prediction using nearby labeled observations.
4 lessons · Distances, votes, feature scales and neighbor selection.
Open K-Nearest NeighborsCombine prior class information with evidence from features.
4 lessons · Counts, smoothing, densities and probability interpretation.
Open Naive BayesExplain a separating boundary and the margin around it.
5 lessons · Margins, C, kernels, scaling and classification limits.
Open Support Vector MachinesFollow a sequence of questions, then see how multiple models can combine or correct predictions.
Trace an observation through splits to a leaf prediction.
6 lessons · Partitions, class mixtures, depth, pruning and evaluation.
Open Decision TreesExplain why different trees can be useful together.
5 lessons · Resampling, aggregation, out-of-bag evidence and interpretation.
Open Random ForestsExplain how successive models correct remaining errors.
6 lessons · Sequential updates, residuals, classification and stage selection.
Open Gradient BoostingBuild from weighted sums to small nonlinear networks; this is an introductory path rather than a general deep-learning curriculum.
Trace a forward pass and explain the role of nonlinear units.
5 lessons · Shapes, activations, gradients, evaluation and limitations.
Open introductory MLPBring everything together: compare candidates on matching data, metrics and information boundaries.
Choose between candidates using permitted development evidence.
6 lessons · Fair comparisons, search, validation and responsible final evaluation.
Open Model Selection & ComparisonOne step-by-step page per model: why use it, the idea drawn, build it from scratch, use scikit-learn, tune it, choose it or not, and where it goes wrong. New to ML projects? Read how to handle a problem end to end. Not sure which model? Start with How to choose a model.
The original EV01 prototype remains available. Its records are separate from the full Evaluation Foundations module.
View supported learning recordsLesson refinement, manual accessibility and real learner validation are ongoing. This page does not change any model, assessment or saved progress.