What are the four common anonymization technique families for big data?

Question

Accepted Answer

Data masking & tokenization, generalization & suppression, noise addition & perturbation, and synthetic data generation.

Technique
What it does

Masking & tokenization
Replace sensitive values with fake-but-realistic data, or with tokens reversible only via a secure key.

Generalization & suppression
Replace specifics with broader categories (exact age → age range) or remove highly identifying attributes entirely.

Noise addition & perturbation
Add controlled statistical noise so exact original values can't be recovered, while overall patterns survive.

Synthetic data generation
Create entirely artificial datasets that match the statistical properties of real data but contain no real personal info.

Tip: Tokenization is reversible (with the key) — so tokenized data is pseudonymized, not truly anonymized. Generalization/suppression and noise lean toward irreversible anonymization.

Technique	What it does
Masking & tokenization	Replace sensitive values with fake-but-realistic data, or with tokens reversible only via a secure key.
Generalization & suppression	Replace specifics with broader categories (exact age → age range) or remove highly identifying attributes entirely.
Noise addition & perturbation	Add controlled statistical noise so exact original values can't be recovered, while overall patterns survive.
Synthetic data generation	Create entirely artificial datasets that match the statistical properties of real data but contain no real personal info.