Machine Learning: Principles and Practice

About this course

The core machine-learning course. Learners master the statistical learning framework and the methods that still do most of the work in industry - linear models, tree ensembles, kernels, clustering, forecasting and ranking - with the discipline that makes results trustworthy: leakage-free validation, metrics chosen from business costs, careful feature engineering, explanation, fairness assessment and model cards a reviewer can act on.

For: student, data professional, software engineer.

Learning outcomes

On successful completion you will be able to:

1. 1Explain the statistical learning framework, including generalisation, the bias-variance trade-off, regularisation and the assumptions behind common supervised and unsupervised methods.
2. 2Build supervised models (linear and logistic models, tree ensembles and kernel methods) with scikit-learn and gradient-boosting libraries, using pipelines that prevent leakage.
3. 3Design validation schemes and choose metrics that match the business problem, including for imbalanced, grouped and time-ordered data.
4. 4Engineer features and handle missing data, categorical variables and dataset shift correctly.
5. 5Apply clustering, dimensionality reduction and anomaly detection, and interpret their results critically.
6. 6Interpret model behaviour with global and local explanation methods, and assess fairness across groups.
7. 7Take responsibility for an end-to-end modelling project, documenting data, assumptions, limitations and risks in a model card a reviewer can act on.

Modules and lessons

1. Week 1The learning problem⌄

   • The statistical learning framework
   • Validation design and choosing metricslab
2. Week 2Linear models⌄

   • Linear regression and regularisation
   • Logistic regression and generalised linear modelslab
3. Week 3Trees and ensembles⌄

   • Decision trees
   • Bagging, random forests and gradient boostinglab
4. Week 4Features and data⌄

   • Feature engineering
   • Missing data, leakage and dataset shiftlab
5. Week 5Kernels, Bayesian methods and tuning⌄

   • Kernel methods and support vector machines
   • Bayesian methods and hyperparameter optimisation
6. Week 6Unsupervised learning⌄

   • Clusteringlab
   • Dimensionality reduction and anomaly detection
7. Week 7Forecasting and ranking⌄

   • Time-series forecasting
   • Recommendation and ranking
8. Week 8Trustworthy models⌄

   • Interpreting and explaining modelslab
   • Fairness, documentation and model cards

Assessment

• Quiz (15%)

  Eight weekly quizzes drawn from the lesson quiz banks; best seven count.

• Lab (30%)

  Six notebook labs with automated checks and short written interpretations.

• Project (40%)

  End-to-end modelling project on a provided business dataset, with a justified validation design, explanation and fairness analysis, and a model card.

• Viva (15%)

  Twenty-minute technical interview on the project, held live on camera.

Pass mark 60%. Grades: Distinction, Merit, Pass.