Features, explained simply
A feature is a measurable detail about an example: its colour, size, weight, number of legs or how round it is. A model can only learn from the features it is given.
Choosing features is a big part of building a good model. If you want to tell apples from bananas, shape is a strong clue. Colour might mislead you, because green apples and green bananas both exist.
Two models trained on the same examples but with different features can make different predictions. When a model gets something wrong, one of the first questions to ask is: which clues was it using?
Example: Two clue detectives
- Detective A only looks at colour. Detective B only looks at shape.
- A blue berry arrives. Detective A has only seen red berries, so it guesses "crystal".
- Detective B sees something small and round, so it guesses "berry".
- Same data, different features, different answers.
Try this
- Pick two animals and list five features you could measure. Which one feature best tells them apart?
- Play twenty questions and notice which questions are the most useful features.
- Find a feature that seems useful but would fool a model, like colour for ripe and unripe fruit.