Skip to content
BENIBIX Learn

Feature Engineering

Also known as Feature Creation, Feature Extraction (Manual)

You make new, better inputs from raw data. Then the model learns with ease.

Think of it like

It is like cutting fruit before you give it to a baby. The same food, made easy to take in.

Example

A shop's data holds the time of each order. Lucy adds a column for "weekend or not". Her sales model gets far better at guessing busy days.

Why it matters

Better inputs often help more than a fancier model. A clear clue is easy to learn from.

Definition

Feature engineering is making new inputs from raw data, so a model can learn more easily. You mix, split or count the columns you already have. A date can become "day of the week". Price and size can become "price per square metre". Good features put the clue the model needs in one place. Deep learning finds many features on its own, so classic models gain the most.
Price per square metre is a new column made from two old ones.

Common mistake

Building a feature from facts you will not have when the model is used. That leaks the answer into training.

Key takeaways

  1. 1
    From raw columns. Mix, split or count what you have.
  2. 2
    Carry the clue. Put the useful signal in one place.
  3. 3
    Only real inputs. Use facts known at prediction time.

Learn it properly