A feature is any measurable input a model uses to make a prediction.
DefinitionWhat it means
A feature is an individual input variable fed into a model, such as a customer's age, a pixel value, or a word count, that the model uses to learn patterns and make predictions. In classical machine learning, data scientists hand-engineer features through a process called feature engineering, selecting and transforming raw data into signals the model can use; deep learning models, by contrast, learn useful features automatically from raw data like text or images.
Why it mattersWhy you should care
Feature quality is often the single biggest driver of model performance in traditional ML systems, more so than model choice, which is why data science teams spend most of their time on feature engineering and feature stores rather than model tuning. Understanding features also clarifies why LLMs feel different: they largely remove the manual feature-engineering burden by learning representations directly from raw text.
At a glanceSee it
The transform a feature needs is decided by its type — numbers get scaled, categories encoded, and text and time each get their own treatment.
A feature store computes each transform once and feeds both training and live serving, so any divergence between the two paths surfaces as train-serve skew.
Where you see itIn the wild
- Feature stores in production ML platforms
- Feature importance charts in a model evaluation report
- Design reviews weighing how to engineer features for a fraud model