Big idea. A predictive model learns a rule from past examples to estimate something not yet known. Its worth is judged by comparison with a simple baseline on data it has not seen.
Outcomes served. O1 (describe supervised learning) and O2 (measure error and compare with a baseline).
Plan your six hours
| Part | Minutes |
|---|---|
| Opener and orientation (this page) | 15 |
| M1.L1 What a predictive model is | 55 |
| M1.L2 Features, labels and tables | 50 |
| M1.L3 Error and the baseline | 50 |
| M1.L4 Train, test and over-fitting | 50 |
| Practice set | 35 |
| Lab: Baseline, rule and test | 50 |
| Mastery checkpoint | 30 |
| Discussion prompt and recap | 25 |
| Total | 360 (6 hours) |
Assumed. Nothing from earlier modules. You should be comfortable with percentages and averages.
Where this topic appears in practice. Anyone who reads a claim that a model “predicts” something, whether customer churn, house prices or disease risk, needs the same questions: what are the features, what is the label, what is the baseline, and was it tested on unseen data? See the entry on machine learning.
Privacy. Your answers stay with you. This course does not store, collect or send anything you write or tick.