Modelling covers feature engineering, algorithm selection, training candidate models and selecting among them without touching the test data.
What the base framework establishes
The clause describes the step that produces a trained and tested model ready for deployment. It covers feature engineering, the quantitative descriptors the model sees; the selection of an algorithm; the training of one or more candidates; and the selection of the best candidate against the design objectives and the risks identified during development. The framework is explicit that model selection belongs to this stage and must stay separate from evaluation on test data. The public summary of the pipeline names iterative and non-sequential ordering, because analysis can change features, features can change the algorithm choice, and the algorithm can change the labelling need.
Why it matters to an AI management system
The separation between selection and final evaluation is the integrity rule of the pipeline. When the test set is used to choose a model, the reported result is no longer an estimate of unseen performance. A governance record that cannot show the separation cannot defend the number.
The public draft summary names no change to 8.4 Modelling. The final text could still touch it.
Evidence a reviewer samples
| Category | Artefact |
|---|---|
| Feature record | Each feature, its construction and its source. |
| Candidate comparison | The candidates trained, the selection criterion and the validation result. |
| Test-set holdout | Evidence that the test set was untouched during selection. |
One question to prepare for. Was the test set used to choose this model?
Common gaps
- Features are created in code with no record of their definition.
- Candidates are compared on the test set and the best is reported.
- The selection criterion changes after the results are seen.
Prepare before publication
Record the feature definitions and the candidate comparison. Lock the test set before selection and record the lock.
Dates that matter
No separate date attaches to 8.4; the record dates are the stage date, the planning estimate and the CEN expectation stated above.
Cross-references
6.3 Model · 8.3 Data preparation · 8.5 Verification and validation · 8.6 Model deployment
Source
Source for 8.4: ISO/IEC 23053:2022, clause 8.4. The public sample stops before 8.4, so this page relies on the public contents and the public summary and does not quote the clause. The two official records are linked on this page so any line can be checked.
Not legal advice; does not certify. A reading of the framework and the evidence a reviewer commonly samples, not a substitute for the standard or your auditor's rules.