Predictive Modelling with Machine Learning

Introduction

Please note that this is a preliminary course description. The final version will be published in June 2026.

The vast amount of widely available data allows machines to solve challenging tasks without explicitly being programmed to do so, often outperforming existing methods based on domain knowledge and/or human experts.

In this course we focus on basic machine learning methods, both for supervised and unsupervised learning. We will look at how exactly machines "learn" from the data, and how they use the knowledge learned during training to solve tasks of interest.

The course covers both the theory (looking at the methods more rigorously than introductory machine learning courses) and the practice of machine learning (using Python).

Course content

• Statistical foundations: recap of probability & inference; bias/variance trade‑off; over‑fitting.
• Linear & penalised regression: ridge, lasso, elastic‑net; diagnostics.
• Generalised linear models: link functions; exponential‑family distributions; logistic & Poisson regression.
• Classification algorithms: k‑NN, naïve Bayes, support‑vector machines, tree‑based models.
• Clustering: k‑means, hierarchical, density‑based methods.
• Neural networks: feed‑forward architectures, back‑propagation, optimisation, regularisation, drop‑out.
• Ensembles: bagging, random forests, gradient boosting (XGBoost).
• Model assessment & selection: cross‑validation, bootstrap, information criteria, learning curves.
• Responsible ML: fairness, interpretability, reproducibility.