What are the common methods for generating statistically synthetic data?

CART decision trees, parametric models, GANs, VAEs, and Bayesian networks — trading simplicity/interpretability against ability to capture complex patterns.

Five common approaches:

• CART (Classification & Regression Trees) — decision trees model relationships; simple and interpretable but may miss complex patterns.
• Parametric models — assume specific statistical distributions; fast and efficient, but only if the distributional assumptions hold.
• GANs (Generative Adversarial Networks) — two neural nets in competition; capture complex patterns but need significant compute.
• VAEs (Variational Autoencoders) — compress data into a latent space and reconstruct; good for high-dimensional data but complex to tune.
• Bayesian networks — model probabilistic relationships between variables; interpretable structure but computationally intensive at scale.

Underneath, generation happens via rule-based (predefined business logic), statistical modeling (match the original's statistical properties), or ML models (learn complex patterns, then generate).

Tip: GANs/VAEs = power at high compute cost; CART/Bayesian = interpretability; parametric = speed if your distribution assumptions are right.