Skip to content
ML Academy

Engineering, production, governance, cost, communication · insight-0278

Minimise personal data in features and give training snapshots a retention period

Evidence type
documented fact
Difficulty
medium
Roles
ML engineer, Data scientist, Data analyst
Industries
telecom, banking, healthcare

Summary: situation, action, reason, boundary

  1. Situation

    A churn model's training set copies full customer records, including fields the model never uses, and snapshots are kept indefinitely.

  2. Action

    Select only the fields that features need, pseudonymise identifiers, store derived features rather than raw records where possible, and set and enforce retention periods for training snapshots and logs with your privacy team.

  3. Reason

    Under GDPR, personal data should be limited to what is necessary for the purpose and kept identifiable no longer than necessary.

  4. Boundary

    Any model trained on personal data, especially in the EU and similar regimes.

Symptom or risk

Unused personal data increases breach impact and compliance exposure without improving the model.

Exceptions

Archiving for research or statistics may have longer retention with safeguards; follow legal advice, not this card.

Trade-offs

Short retention limits reproducing old models; keep model artifacts and metadata, and document what was deleted.

Pitfalls

Pseudonymised data treated as anonymous when re-identification is still possible.

Lessons and concepts

day 53day 19
PrivacyData contracts

Claims and their sources