Project exam · Data Science, Machine Learning and AI
Project exam: a crop adviser with k-nearest neighbours
The brief
In 75 minutes, build the model behind a farmers' helpline: read the field readings while skipping bad rows, scale the features, measure distance, predict a crop by k-nearest neighbours, score it on held-out farms and pick k by leave-one-out validation.
Why it matters
Agritech start-ups and state agriculture departments give crop and fertiliser advice from soil and weather readings, and k-NN is a common first model because it is simple to explain to a farmer: "farms most like yours grew this". The exam checks the skills a junior ML job needs: clean input, scaling, a correct model, and evaluation that does not cheat.
Skills you practise
- reading CSV safely
- min-max scaling
- Euclidean distance
- k-nearest neighbours
- test accuracy
- leave-one-out validation
What you will have at the end
adviser.py: a k-NN crop adviser that reads messy field data, scales it, predicts a crop and reports honest test and validation accuracy.
How it is marked (rubric, 100 points)
| Correct modelEvery step passes: reading, scaling, distance, prediction with ties, test accuracy and leave-one-out. | 50 |
|---|---|
| No leakageThe scaler and k are chosen from training data only; the test set is used just to report the final score. | 20 |
| Clean functionsEach function does one job and reuses the earlier ones; no repeated code. | 20 |
| Finished in timeAll six steps handed in within 75 minutes. | 10 |
- 6tasks
- 75 mintime limit
- 60%to pass
Rules: a project exam
- The exam opens full screen. Stay on this page until you hand it in.
- Opening another tab or app, or leaving full screen, counts as one time away. After 3 times the exam ends.
- No copy or paste during the exam.
- No hints and no answers. You can run your code, but the checks run only when you hand in.
- Every task is checked automatically when you hand in, or when the time is up.
- You see which tasks passed at the end. You can take it again.