Assessment Guidelines

This document describes how you are assessed in MST0052, what is expected at each grade level, and how the semester project and oral exam each contribute to the final course grade. Read it carefully early in the semester — it is designed to help you plan your project and your exam preparation.

Overview

Component	Weight	Format
Semester project	30%	Individual written report (PDF)
Oral exam	70%	Closed book, no notes, no tools

The two components have distinct purposes. The semester project assesses the quality of the work you submit. The oral exam assesses your understanding of your submitted work and your command of the core course material.

To pass the course, the student must obtain a passing grade in both components.

Part 1: Semester project (30%)

Format

Individual work. Each student submits their own project and report.
Deliverable: a single PDF report. Code is not submitted and not graded.
Sharing your code is encouraged, not required. Putting your project on GitHub is good practice and useful beyond this course, but it plays no part in the grade.
Deadline: 20 November 2026 at 12:00.
Scope: a complete predictive modelling project — problem definition, data, preprocessing, modelling, evaluation, and discussion.

What a good report contains

A typical report includes:

Problem statement — what question are you answering, and why does it matter?
Data — source, structure, preprocessing choices, and any issues you encountered.
Methodology — which models you used, why, and how you validated them.
Results — evaluation metrics, comparisons between models, and honest discussion of limitations.
Reflection — what you learned, what you would do differently, and any ethical or practical considerations.
AI usage statement — a brief note on which AI tools you used and how.

There is no strict page limit. Write as much as is needed to make your work clear and no more. Quality beats length.

Collaboration policy

Discussion and collaboration are encouraged. Talk to other students, compare approaches, help each other debug, share ideas.
The final report must be your own independent work. You must write the text, do the analysis, and make the modelling decisions yourself. Copying reports, or submitting work built as a joint effort, is not allowed.

AI usage policy

You are explicitly allowed and encouraged to use generative AI tools (Gemini, ChatGPT, Claude, Copilot, Cursor, etc.) throughout the project. There are no restrictions on what you can use AI for — including writing code, drafting text, exploring advanced methods, building figures, or learning new topics.

Use AI to go further than you could on your own. Dig into techniques we did not cover in class. Try methods from papers. Build something you are proud of. I hope to learn something interesting from reading your projects.

The one requirement: anything in your report must be something you genuinely understand and can defend on the oral exam. This is the real constraint, and it is explained in the next section.

Feedback opportunities during the semester

There are two structured feedback points in the schedule, plus open feedback channels throughout:

Lecture 8 (optional project showcase): volunteer presenters get feedback from the instructor in front of the class, and the audience gets ideas from what other students are trying. Sign up in Canvas by 6 October; everyone is welcome to attend.
Lecture 15 (project workshop): rehearse with peers before the final submission.
Office hours, email, or a requested meeting: the right channel for individual feedback at any point. If you want my opinion on your project and don’t want to present at L8, just ask — any week.

Use these channels. They are the best way to catch methodology issues (leakage, wrong metrics, missing baselines) while you still have time to fix them.

Part 2: Oral exam (70%)

Format

Timing: a few weeks after the project submission deadline, in December 2026 (exact dates TBA).
Closed book. No notes, no laptop, no phone, no AI tools, no printed report. You bring yourself.
Duration: announced closer to the exam.
Structure: two parts, in this order.

Part A: Project defence

The examiner has read your report. Expect questions such as:

Why did you choose this dataset and this problem?
Why did you use method X rather than method Y?
How did you validate your model? Why this validation strategy?
What does this specific number in Table 3 mean, and how was it computed?
What is the biggest weakness of your analysis?
If you had more time, what would you do next?
Walk me through what is happening inside the model you used here.

You should be able to explain every choice, method, figure, and number in your report. If you used a technique, you need to be able to describe what it does, why it fits your problem, and what its limitations are. This applies equally to things you learned with AI assistance — the oral exam is where you show that the learning actually happened.

Part B: General course questions

A smaller part of the exam covers the general syllabus at a conceptual level. These are overall, not deep questions — the kind of thing any student who followed the course should be able to answer. Examples:

What is the difference between linear and logistic regression?
What is the difference between a decision tree and a random forest?
What does regularisation do, and why might we want it?
When would you use unsupervised learning rather than supervised learning?
What is the bias-variance tradeoff, in one or two sentences?

You do not need to memorise formulas or derive anything on the board. You need to show that you understand the core ideas and can compare the main methods.

How the two parts interact

The semester project and the oral exam are separate assessment components.

The semester project assesses the quality of the submitted analysis and report.
The oral exam assesses your ability to explain, justify, and reflect on the submitted work, and to demonstrate conceptual understanding of the course syllabus.

The oral exam may use your project as its starting point, but it does not re-grade the written submission. The two components are graded separately and then combined using the stated weights, provided that the student has obtained a passing grade in both components.

Students who use more advanced methods should be able to explain those methods at an appropriate level. However, all students are assessed against the same overall standards: correctness, clarity of reasoning, understanding of limitations, and command of relevant course concepts.

Grade descriptors

BI uses the ECTS A—F scale. The descriptors below apply to the final course grade, based on separate performance in the project and oral exam, with weights of 30% and 70% respectively. A passing result in both components is required in order to pass the course.

A — Excellent

Project: ambitious, well-scoped problem. Methodology is rigorous and often extends beyond the core syllabus (advanced methods, careful validation, thoughtful feature engineering, or a genuinely interesting dataset). Report is clear, describes the pipeline in enough detail that it could be reproduced, and shows independent thinking.
Oral defence: confident and precise explanations of every modelling choice. Handles follow-up questions well. Acknowledges limitations honestly and suggests meaningful extensions.
General questions: fluent conceptual understanding across the syllabus.

B — Very good

Project: solid application of course methods, with at least one aspect that goes beyond the basics. Report is well-structured and documents the pipeline clearly enough to follow.
Oral defence: clear explanations of core choices, minor hesitation on the most advanced aspects, but no major gaps in understanding.
General questions: good grasp of the main concepts and tradeoffs.

C — Good

Project: correct application of standard course methods to a reasonable problem. Report is competent but not especially deep. Minor methodological issues possible.
Oral defence: can defend the core of the project. Some follow-up questions expose weaker understanding, but the main analysis is understood.
General questions: adequate understanding of the main syllabus topics.

D — Satisfactory

Project: works end-to-end but has noticeable gaps (weak baselines, questionable validation, thin evaluation, or unclear reasoning).
Oral defence: partial understanding of own work. Struggles to explain choices beyond a surface description.
General questions: basic understanding with visible gaps.

E — Sufficient

Project: minimal viable submission. Methodology flaws are more than cosmetic.
Oral defence: significant gaps in understanding the submitted work.
General questions: only the most basic concepts can be explained.

F — Fail

Any of the following will typically result in a failing grade in a component:

Project is missing major components or is not submitted.
Student cannot meaningfully defend the submitted project — for example, cannot explain the methods used, cannot interpret results, or cannot describe how the analysis was carried out.
Student cannot answer basic syllabus questions at a conceptual level.
Evidence of academic misconduct (see below).

A failing grade in either component results in a failing grade for the course.

Academic integrity

Your report must be your own independent work.
Collaboration on ideas and debugging is fine; joint submissions are not.
Using AI to help you build and learn is fine; submitting AI output you do not understand as if it were your own reasoning is not — and it will show up in the oral exam regardless.
Follow BI’s regulations on academic misconduct.