Skip to content

Teacher guide — Grade 12, Unit 2: Regression and classification

Time: 3 lessons + project (about 5 periods). Track: B (code).

# Lesson Time
1 Regression: predicting a number 40–45 min
2 Classification: predicting a category 40–45 min
3 Choosing and explaining 40–45 min
P Unit project: Explain a prediction 2 periods

Before you start

  • Materials: Notebook notebooks/reg_class.ipynb.
  • Read docs/pedagogy/lesson-anatomy.md and docs/pedagogy/ai-use-policy.md.

If you have no computers / no internet

ML concepts can be taught with cards (Grade 10 Unit 5) and printed confusion matrices. Code lessons need a lab day; the capstone can be done with spreadsheets and a decision rule, but call that 'a simple rule', not 'a model'.

Common misconceptions

  • Logistic regression predicts a class directly (it predicts a probability first).
  • Bigger coefficient = more important if scales differ.

Differentiation

  • Support: give a partially filled table / sentence starters; pair with a data-steward role.
  • Stretch: ask "what would change your conclusion?" and "who is missing?"

Answer key (check questions)

Lesson 1: Regression: predicting a number 1. Regression predicts…
Answer: A number. 2. Coefficient 0.05 for visits means…
Answer: +0.05 kg per extra visit, other features fixed. 3. Same as Grade 11 least squares?
Answer: Yes: same line.

Lesson 2: Classification: predicting a category 1. Probability range?
Answer: 0 to 1. 2. What turns a probability into a class?
Answer: A threshold. 3. Who chooses the threshold?
Answer: People, based on costs of different mistakes.

Lesson 3: Choosing and explaining 1. Predict rainfall in mm: type?
Answer: Regression. 2. Predict 'pass/fail': type?
Answer: Classification. 3. Why favour explainable models?
Answer: People affected deserve to understand decisions.

Assessment

Starter quiz each lesson (3 questions, one recalling the previous lesson); unit quiz in quiz.yml; project marked with the rubric in docs/pedagogy/assessment.md; 2-minute oral check per group.

Alignment

Enrichment aligned to NEB Grade 12 Computer Science (programming, databases, emerging technologies) and Maths/Statistics.

Answers: practice exercises

  1. (a) regression (count; Poisson-type models later) (b) classification (c) regression (d) multi-class classification.
  2. z = 0 → 0.5; z = 2 → 1 ÷ (1 + e^−2) ≈ 0.88; z = −2 → ≈ 0.12.
  3. Odds 3 = p ÷ (1 − p) → p = 3 ÷ 4 = 0.75.
  4. z = −1.593 + 0.345 × 4 + 0.088 × 3 ≈ 0.05 → p ≈ 0.51.
  5. 1.41² ≈ 1.99 (about double the odds); 1.41⁴ ≈ 3.97.
  6. Linear predictions range from about 0.15 to 0.73 on this test set — inside 0–1 here only because the features are limited; with stronger features it would go outside. Logistic regression always gives valid probabilities and models the S-shaped relationship.
  7. MAE 0.334 kg, RMSE 0.422 kg; RMSE squares errors, so a few large misses raise it more.
  8. Depth 2: test accuracy 0.576, 4 leaves — and its rules reduce to "4 or more visits → facility" (the Grade 10 rule!). Depth 10: 0.579 with 35 leaves. Same accuracy; give the health post the depth-2 version.
  9. The model's uncertainty is large; 0.49 and 0.51 are practically the same. Better: a "grey zone" (e.g. 0.4–0.6) reviewed by a person, or decisions based on several pieces of information; always allow appeal.
  10. e.g. "In our records, babies of mothers who had more check-ups during pregnancy, and from better-off families, weighed a little more at birth — about 60 grams more for each extra check-up. But these two things explain only a small part of why babies differ, so we cannot predict any one baby's weight well. The data are invented for teaching."

Lesson 1 Make: quintile 3, 5 visits → ≈ 3.01 kg (± about 0.33 kg typical error). Adding age and belt dummies leaves test MAE at ≈ 0.335 kg — no improvement. Lesson 2: flagged counts at thresholds 0.3/0.4/0.5/0.6 = 34/151/270/343 of 375; for 120 visits, a threshold just under 0.4 (check the exact count).

Real-data exercises: notes

  • R1 Flipper length explains most of the variation in body mass (R² ≈ 0.75); species classification ≈ 0.95+.
  • R2 Nepal typically sits near the decision boundary; a good discussion of what a "probability" means for a country.
  • R3 A farmer spraying expensive pesticide that rain would wash off wants high precision for "no rain".