Syllabus
Course description
MST0052 is an intermediate course in predictive modelling and machine learning. The course focuses on traditional machine-learning methods and the statistical principles behind predictive modelling, before introducing a short conceptual bridge to neural networks. The emphasis is on building reliable models, understanding when to trust them, and communicating results clearly.
Learning objectives
Knowledge
- Explain the statistical foundations of predictive modelling, including overfitting, regularisation, and the bias-variance tradeoff
- Compare core machine-learning methods for regression, classification, clustering, and ensemble learning
- Understand strengths, limitations, and practical tradeoffs related to interpretability, computational cost, and predictive accuracy
- Distinguish traditional statistical learning from deep learning and identify when each is appropriate
Skills
- Design, implement, and validate predictive models in Python using standard scientific and machine-learning libraries
- Build reproducible preprocessing and feature-engineering pipelines
- Apply cross-validation, hyperparameter tuning, and model-selection workflows
- Extend linear models to penalised regression variants (ridge, lasso)
- Evaluate model performance critically and communicate findings in professional reports and presentations
General competence
- Translate practical business or societal problems into data-driven predictive questions
- Plan and execute a substantial semester project with statistical rigour and pragmatic data-science practice
- Reflect on ethical, legal, and sustainability aspects of predictive analytics, including fairness, privacy, and transparency
Assessment
| Component | Weight | Format |
|---|---|---|
| Semester project | 30% | Written report (PDF) |
| Oral exam | 70% | Defend your project + general syllabus questions |
The semester project is the only graded deliverable besides the oral exam.
See the Assessment Guidelines for full details on project requirements, the oral exam format, and grade descriptors (A—F).
Semester project
The semester project is a continuous piece of work running through the entire course. You will:
- Choose a dataset and define a predictive modelling problem
- Build preprocessing pipelines and apply methods as they are introduced in lectures
- Compare models and justify your choices
- Submit a written report (PDF) by 20 November 2026 at 12:00 — code is not handed in, though sharing it on GitHub is encouraged
The lecture schedule includes an optional project showcase at Lecture 8 (volunteer presentations to the class, by signup) and a project workshop at Lecture 15 (peer rehearsal before submission). Both are open to everyone; only the workshop is structured around all students presenting.
See the Project Guidelines for the full scope rules (including the no-deep-learning restriction) and required project elements.
Oral exam
The oral exam (December 2026, exact dates TBA) has two parts:
- Project defence — present your semester project and answer questions about your modelling choices, results, and methodology
- General questions — demonstrate understanding of topics from the syllabus (methods, concepts, tradeoffs)
Teaching format
The course consists of 16 sessions combining lectures with interactive discussion. Students are expected to prepare by reading assigned material and making steady progress on their semester project.
There is no live coding in lectures. Code examples are shown on slides, and you are expected to work through them on your own as you build out your semester project.
AI tools policy
Generative AI tools (ChatGPT, Claude, GitHub Copilot, etc.) may be used as assistants in this course. The following guidelines apply:
- You are responsible for everything you submit. If an AI tool produces incorrect code or analysis, it is your responsibility.
- Disclose your usage. In your project report, note which AI tools you used and how they contributed.
- Use AI to learn, not to bypass learning. AI tools are most valuable when you understand what they produce and can verify it.
- Verify all AI-generated output. Models can hallucinate, produce buggy code, or give misleading statistical advice.
Academic integrity
All submitted work must be your own. You may discuss approaches with classmates, but the code, analysis, and writing in your project must be individually produced. Plagiarism or misrepresentation of AI-generated work as your own understanding will be treated according to BI’s regulations on academic misconduct.