Regularisation
Students learn how shrinkage methods reduce instability and help prediction models handle many predictors.
Module 4
Regularisation, Ensembles and Modern Prediction Models
This module introduces more flexible prediction tools used in medical machine learning: ridge regression, lasso, elastic net, random forests, gradient boosting, support vector machines and responsible model comparison.
Module aim
Use complex models without losing judgement.
The purpose of this module is to explain how modern prediction methods can improve performance while still requiring careful validation, calibration and clinical interpretation.
5
Lessons
R
Coding labs
Preparing
Module status
Modern
Prediction focus
Ensemble learning
The module introduces forests and boosting as methods that combine many simpler models into stronger predictors.
Responsible comparison
Modern models are compared through validation, calibration, usefulness and interpretability rather than accuracy alone.
Module lessons
Move from regularisation to modern model comparison.
Each lesson introduces a more flexible prediction idea while keeping the same biostatistical standard: validate honestly, avoid overclaiming, check calibration and explain why the model is useful.
4.1
Lesson 4.1
Ridge, lasso and elastic net
Learn how regularisation controls model complexity, shrinks coefficients and helps prediction models behave better with many or correlated predictors.
Open lesson →
4.2
Lesson 4.2
Random forests
Understand random forests as ensemble tree models that combine many decision trees to improve prediction stability and reduce overfitting.
Open lesson →
4.3
Lesson 4.3
Gradient boosting
Study boosting as a sequential learning approach where models are built stage by stage to improve prediction performance.
Open lesson →
4.4
Lesson 4.4
Support vector machines and flexible boundaries
Learn how support vector machines create separating boundaries and why flexible decision boundaries need careful validation in health data.
Open lesson →
4.5
Lesson 4.5
Comparing models responsibly
Bring modern prediction models together by comparing them through validation, calibration, clinical usefulness, interpretability and reporting discipline.
Open lesson →
Learning route
Complete model comparison before applied case studies.
Module 5 assumes that students understand classical prediction models, validation, performance metrics and modern model comparison before applying them to realistic health-data projects.
Continue to Module 5 →Course pathway
Return to the full ML in Biostatistics course.
Use the course homepage to review the full pathway, case studies, datasets, scripts and supporting course resources.
Back to course →