This page is due for review. Re-read the key points and answer the self-check questions before recording your video.
ML/DL · Hub
ML foundations
Eight ideas every machine-learning project rests on: the learning paradigms, how to score classifiers and regressors, why models underfit or overfit, how regularization helps, how to evaluate honestly with cross-validation, how to prepare data, and how to avoid leakage.
Why it matters: most failed ML projects fail here, not in the choice of algorithm: a wrong metric, an overfit model, unscaled features or a leaky evaluation.
5 random questions mixed from all eight foundation pages.
Mixed review
Answer each one out loud before opening the sample answer.
flowchart TD
A{"Where are you in the project?"} -->|"framing the problem"| P["Learning paradigms"]
A -->|"preparing data"| PR["Preprocessing"]
PR --> DL["check for data leakage"]
A -->|"choosing how to score"| M{"Category or number?"}
M -->|category| CM["Classification metrics"]
M -->|number| RM["Regression metrics"]
A -->|"model fits badly"| BV["Bias-variance"]
BV -->|overfitting| RG["Regularization"]
A -->|"comparing models or settings"| CV["Cross-validation"]
Key points
The paradigm comes first because labels, or their absence, decide which methods are possible.
Metrics must match costs because a missed positive and a false alarm are rarely equally bad.
Training error is not enough because flexible models can memorise noise.
Cross-validation gives honest comparisons because every score comes from unseen data.
Everything is fitted on training data only because anything else leaks information into the evaluation.