Features and Labels
Every supervised dataset is built from two parts: features and labels. Getting this vocabulary right is the foundation for everything else.
A feature is a measurable property of an example. For a house, features might be square footage, number of bedrooms, and age. Each example is described by a vector of feature values. A label is the target you want the model to predict, such as the sale price.
Some useful terms:
- Feature vector is the full set of inputs for one example
- Target is another word for the label
- Feature engineering is the craft of turning raw data into informative features
The quality of features often matters more than the choice of algorithm. A model can only learn from the information you give it. Irrelevant features add noise, while a well chosen feature can make a hard problem easy.
Labels must be accurate and consistent. Noisy or mislabeled targets cap how well any model can do. In many real projects, most of the effort goes into preparing clean features and trustworthy labels rather than tuning the model itself.
Key idea
Features are the measured inputs and the label is the target answer; good features and clean labels matter more than the algorithm.