Frame
Define the prediction target, horizon, unit and intended decision.
SRAI Book 2 · Chapter 1 · PU-B02-C01
Supervised Learning Foundations
Learn how labelled data becomes defensible prediction evidence through valid evaluation, leakage control, credible baselines, uncertainty and accountable use.
01 / LEARNING OUTCOMES
Turn labelled data into defensible prediction evidence.
- Define the observational unit, target, features and prediction horizon.
- Explain empirical risk, expected risk and generalization.
- Design training, validation and test partitions that reproduce intended use.
- Detect target, temporal, group and pipeline leakage.
- Compare learned models with credible baselines.
- Interpret regression, classification, calibration and subgroup evidence.
02 / VALIDATION SEQUENCE
Frame, partition, compare and govern.
Partition
Separate training, validation and test evidence without leakage.
Compare
Evaluate learned models against credible operational baselines.
Govern
Record uncertainty, limitations, ownership and acceptable use.
03 / CONTROLLED RESULTS
A strong metric is useful only when the evaluation design is valid.
The canonical notebook predicts next-day service demand using a chronological test design and reproducible Python workflow.
A post-outcome feature produces an apparently excellent result, but it is explicitly rejected because it leaks unavailable future information.
VERIFIED NOTEBOOK RESULTS
| Regression RMSE | 30.876 |
| Seasonal baseline | 40.496 |
| Training-mean baseline | 103.958 |
| Classification ROC AUC | 0.955 |
| Rejected leakage RMSE | 2.048 |
A model can produce an impressive score.
The score alone does not prove that the model will generalize to its intended operational setting.
Evidence must respect the prediction-time boundary.
Valid partitions, credible baselines and leakage controls make the performance claim defensible.
04 / SRAI DECISION STANDARD
Generalization is a controlled claim—not merely a score.
- The information available at prediction time is explicit.
- The evaluation design reproduces intended use.
- Learned models are compared with credible baselines.
- Leakage demonstrations are identified and rejected.
- Uncertainty, limitations and accountable use remain visible.
CONTROLLED RESOURCES
Read, reproduce, practise and review.
READ
View the complete GitHub production unit ↗
Audited Chapter 1
Controlled publication edition.
PDF ↗ REPRODUCECanonical notebook
Executable and independently verified.
IPYNB ↗ PRACTISEStudent assessment
Progress toward independent application.
PDF ↗ REVIEWInstructor solutions
Review the explained answers and marking guidance.
PDF ↗ APPLYExecutive brief
Connect analytical controls to decisions.
PDF ↗