MST0052
## MST0052 -- Lecture 15 ### Project Workshop Fall 2026 --- ## Today's plan This is a working session, not a traditional lecture. 1. **Present** your project to a small group (5-8 minutes) 2. **Receive feedback** from peers and instructor 3. **Identify** what to improve before submission --- ## What to present - **Problem:** What are you predicting and why? - **Data:** What dataset, how many observations, key features - **Pipeline:** Preprocessing, feature engineering, model choices - **Results:** Best model, key metrics, what you learned - **Limitations:** What didn't work, what you would do differently Keep it concise. This is a rehearsal, not a defence. --- ## Giving useful feedback When you listen to a peer's presentation, consider: - Is the problem clearly defined? - Does the preprocessing make sense for the chosen models? - Are the evaluation metrics appropriate? - Are the results interpreted honestly (not just "my model got 95%")? - What would make the report stronger? Be specific and constructive. --- ## Common pitfalls to check - **Leakage:** preprocessing fitted on full data before splitting - **Wrong metric:** accuracy on imbalanced data - **No baseline:** jumping to complex models without a simple comparison - **Missing documentation:** no description of choices or AI usage - **Overfitting the test set:** tuning on test data instead of validation --- ## Project submission checklist - [ ] Clear problem statement and motivation - [ ] Documented preprocessing pipeline - [ ] At least 2-3 models compared fairly (same CV, same metric) - [ ] Results table with key metrics - [ ] Discussion of limitations and tradeoffs - [ ] Reproducible code (someone else can run it) - [ ] AI-use statement - [ ] Proofread report --- ## Preparing for the oral exam The oral exam (December) covers: 1. **Your project** -- be ready to explain every choice 2. **General syllabus** -- methods, concepts, tradeoffs from the course Good preparation: - Can you explain **why** you chose each model? - Can you explain the **bias-variance tradeoff** in your project? - Can you describe what would change if the dataset were different? --- ## Summary - Use today to get feedback and improve your project - Focus on honest evaluation and clear communication - Check for common pitfalls (leakage, wrong metrics, missing baselines) - Start preparing for the oral exam now --- ## What's next **Lecture 16:** Summing up and Q&A - Course recap - Method comparison - Exam preparation