AI Glossary Term
What does “Feature Extraction” mean?
Converting raw data into informative features that a model can use.
How Feature Extraction works
Example
To recognize spoken commands, audio is first converted into a spectrogram showing which frequencies are present over time. A model then works with these features rather than millions of raw samples.
Why it matters
Good features shrink and clarify the data, making learning faster and more reliable. Pre-trained networks are often used as feature extractors for new tasks.
Common questions
How is it different from feature engineering?
Feature extraction usually means automatically deriving features from raw data. Feature engineering is the wider craft of designing inputs, including by hand.
Can I reuse a pre-trained model for it?
Yes. Taking an image model's internal vectors as features for your own classifier is a common, low-cost approach.
Does extraction lose information?
It can. Choosing features means deciding what to keep, so poor choices may discard useful signal.