AI Glossary Term
What does “Feature Store” mean?
Definition
A managed system for storing and serving validated ML features consistently for training and inference.
How Feature Store works
Example
A fraud team defines 'average spend over 30 days' once in a feature store. Both the training job and the live scoring service read it, so they calculate it the same way.
Why it matters
Mismatched feature calculations between training and production can quietly hurt accuracy. A shared store keeps them consistent and reusable.
Common questions
Do small teams need one?
Often not. It pays off when many models share features or production consistency becomes hard.
Offline vs online store?
The offline store holds historical data for training; the online store serves the latest values quickly for predictions.
What problem does it solve?
Training-serving skew, where features differ between training and production.