An ML dataset includes personal identifiers not required for the prediction task. Which two preparation measures are appropriate? Select 2 answers.

1. Restrict dataset access and apply suitable encryption and retention controls.
2. Minimize or transform unnecessary sensitive fields according to the approved data-use policy.
3. Copy all raw identifiers into public training logs.
4. Assume a model accuracy improvement authorizes any use of personal data.
5. Delete provenance so reviewers cannot identify the sensitive sources.