Data
Features are the individual measurable properties of an example that a model uses as input, such as a house's size, age, and location when predicting its price.
A model cannot look at 'a house.' It looks at numbers describing the house: square footage, number of bedrooms, year built, distance to a station. Each of those is a feature. Choosing which features to include, and how to express them, has a huge effect on how well a model works.
In classic machine learning, humans pick and craft features by hand, a skill called feature engineering. Deep learning changed this. A neural network fed raw pixels or raw text learns its own features in its hidden layers, discovering edges, shapes, or word meanings without being told.
Features can carry hidden problems. A postcode can act as a stand-in for race or income, letting bias in through the back door. Good practice means asking not just whether a feature improves accuracy, but whether it is fair to use.
To predict whether a customer will cancel, a model might use features like months subscribed, number of support calls, and days since last login.