BitWithBite
AI & Machine Learning Worksheet
🤖 Model Evaluation & Tuning
SCORE___ / 20
NameClassDate
After this worksheet you can
Train/validation/test splitsCross-validationPrecision/recall/F1Grid vs random search📚 Quick Recap
A model is only as good as how you measure it. Evaluation done wrong costs companies money — done right, it builds trust. This module covers every tool for squeezing real performance out of your models.
🧠 Section A · Concept Check ● BEGINNER 4 × 1 = 4
1The test set is used:
2K-fold CV averages scores over:
3Precision measures:
4Recall matters most when:
🧮 Section B · Problem Solving ● INTERMEDIATE 2 + 3×3 = 11
5F1 is the harmonic mean of precision and .
6Accuracy misleads on classes.
799% accuracy on 1% fraud data - why bad?
8Explain 5-fold CV.
9Precision vs recall trade-off example?
🚀 Section C · Challenge ● CHALLENGE 5
10Why never tune on the test set?
💭 Reflection — the most useful thing I learned:
A ___/4 B ___/11 C ___/5 Total ___/20
Teacher's Signature
Parent's Signature
✂ answer key — fold or cut before handing out
1-A 2-A 3-A 4-A | 5 recall 6 imbalanced 7 = Predicting all-negative gives 99% - use precision/recall 8 = Split into 5; train on 4, validate on 1, rotate; average 9 = Spam filter: high precision avoids losing real mail; disease test: high recall avoids missed cases | 10 = You'd overfit evaluation - scores stop reflecting reality