Synthetic Data

Synthetic data is data created by algorithms, simulations or generative models rather than collected from real events. It is designed to resemble the structure and statistics of real data without containing the actual records.

Why it matters

It helps when real data is scarce, expensive, imbalanced or sensitive. Teams use it to test software without exposing customer records, to add rare cases such as fraud examples to a training set and to train models on simulated environments. It is not automatically private: a poor generator can leak or reproduce real records, and it can only reflect patterns present in its source, so it may amplify bias or miss real-world edge cases. Models trained heavily on generated data can also degrade over repeated generations.

Example

Real table: 5 fraud rows in 100,000 transactions
Generate 2,000 plausible fraud-like transactions
Train on real + synthetic, then evaluate only on real held-out data

Evaluating on real data is essential to know whether it helped.

Common mistakes

• Assuming synthetic means anonymous and skipping a privacy review.
• Measuring success on synthetic test data only.
• Generating unrealistic values that break business rules.

Related

• Training data
• Dataset
• GAN
• PII

This page is general information written by our editorial team and updated when documentation changes. Servers and providers differ, so test changes safely first. See our editorial policy and disclaimer.