Module 4 · Machine Learning in Biostatistics

Regularisation, ensembles and modern prediction models.

This module moves from transparent baseline prediction to more flexible models. Students learn how penalties, forests, boosting and support vector machines can improve prediction, but also why tuning, validation, calibration and reporting discipline become even more important.

View Module 4 lessons →Open Lesson 1.1 Start learning

5

Lessons

0

Open now

R

Labs planned

July 2026

Full release

What this module builds

Complexity with control.

Complexity needs control

Flexible models can fit richer patterns, but they can also chase noise. This module teaches how regularisation and validation control complexity.

Regularisation stabilises prediction

Ridge, lasso and elastic net add penalties to reduce unstable coefficients, manage collinearity and improve performance on unseen patients.

Ensembles combine many weak patterns

Random forests and boosting use many trees to improve prediction, but their outputs must still be validated and interpreted cautiously.

Tuning can overfit too

Choosing hyperparameters using the wrong data can leak information and make model performance look better than it really is.

The best model is not only the highest AUC

Clinical model comparison should consider calibration, sensitivity, specificity, uncertainty, complexity, explainability and clinical usefulness.

By the end

Students should compare modern models responsibly.

Explain why regularisation is needed in prediction modelling.

Describe the difference between ridge, lasso and elastic net.

Understand coefficient shrinkage and variable selection.

Explain how random forests combine many decision trees.

Describe how gradient boosting learns sequentially.

Recognise why flexible models require careful tuning.

Compare models using validation rather than training fit.

Write cautious reports about modern prediction models.

Module pathway

From regularisation to responsible model comparison.

This module shows how modern prediction models are trained, tuned, validated and reported. The focus is not only on higher performance, but on knowing when extra complexity is justified.

Step 1

Baseline

Begin with a transparent reference model so modern methods are compared against something interpretable.

Step 2

Penalise

Use ridge, lasso or elastic net penalties to reduce instability, shrink coefficients and manage high-dimensional predictor sets.

Step 3

Ensemble

Use forests and boosting to combine many simple learners into a stronger prediction system.

Step 4

Tune

Choose tuning parameters using validation procedures rather than the final test set.

Step 5

Compare

Compare models using discrimination, calibration, sensitivity, specificity and clinical context.

Step 6

Report

Explain model performance, uncertainty, limitations and why the selected model is appropriate for the clinical question.

Lesson design

The modern modelling lessons will be validation-led.

Clinical prediction examples using regularised and ensemble models

Detailed notes linking modern ML methods to overfitting and validation

Browser R coding labs planned for every full lesson

Downloadable R scripts for local model fitting and tuning

Coefficient path, variable importance and performance outputs

Report sections translating model comparison into cautious interpretation

Caution boxes for tuning leakage, overfitting and black-box overclaiming

Current release state

Module 4 is open for preview.

Students can see the Module 4 pathway now. The full lessons are locked while they are redesigned with R scripts, coefficient-path visuals, tuning outputs, variable-importance summaries, validation reports and clinical interpretation.

Compare baseline models, penalised models and ensemble models using validation, calibration, discrimination, uncertainty and clinical interpretability.

ValidationCalibrationResponsible reporting

Locked until July 2026

Start learning →

Join the waitlist

Get access updates when Module 4 opens in July 2026.

Module 4 lessons are currently locked while they are redesigned with regularisation labs, ensemble workflows, tuning outputs, validation comparisons and report-style interpretation.

Module 4 overview stays open.

All Module 4 lessons remain opening in July 2026.

Lesson 1.1 remains available as the course preview.

Waitlist visitors can request early access or release updates.

Waitlist form

Request access.

Recommended start

Begin with the open foundation lesson.

Lesson 1.1 introduces the course structure: prediction question, R script, model output, interpretation, report writing and responsible modelling caution.

Open Lesson 1.1 →