DataTerminology.comNEW
Also known as: Variable, Independent Variable
An individual measurable property of an observation used as model input.
DESCRIPTION
In machine learning, a feature is a measurable attribute derived from raw data to inform a model. Features can be numerical, categorical, text-based, image-based, or engineered combinations. High-quality, informative features materially affect model accuracy and robustness. Feature definitions should be stable, well-documented, and governed within feature stores in production systems. Proper preprocessing—encoding, scaling, and handling missingness—ensures consistent feature behavior across training and inference.
ALTERNATIVE DEFINITIONS
Wikipedia
In machine learning and pattern recognition, a feature is an individual measurable property or characteristic of a phenomenon being observed.
Canonical definition linking to pattern recognition origins.
Google Developers ML Glossary
A measurable property or characteristic of a phenomenon being observed.
Concise ML glossary phrasing emphasizing measurability.
RELATED TERMS
Feature Engineering Feature Extraction Feature Reduction Feature Scaling Feature Store One-Hot Encoding Pipelines Model Input
COMPETING TERMS
Label (Target Variable)
RELATIONS
- Narrower: Rank-Based Feature
KNOWLEDGE AREAS
AI Data Science Data Engineering MLOps
TARGET AUDIENCE
Data Scientist AI Engineer MLOps Engineer Data Engineer Analytics Engineer
EXTERNAL SOURCES
- Wikipedia - Feature (machine learning)
- Google Developers - Guide to creating features
- scikit-learn - Preprocessing and feature transforms
- Feast - Feature store concepts and APIs
TOP QUESTIONS
Why are Features important in machine learning?
Features encode information the model uses to learn patterns. Better features often produce bigger gains than model tweaks, especially with limited data or simpler algorithms.
Category: Relevancy
What makes a good Feature?
Relevance, predictive power, stability over time, low leakage risk, and reasonable cost to compute. Domain knowledge helps craft features aligned to causal mechanisms.
Category: Requirements
How should Features be documented?
Use a feature catalog or store with definitions, owners, transformations, data sources, and quality SLAs. Clear documentation prevents duplication and leakage.
Category: Management
What is Feature leakage?
Leakage occurs when features include information unavailable at prediction time or directly encode labels, causing overly optimistic training performance and poor generalization.
Category: Risk
How do Features differ from Labels?
Features are inputs; labels are outputs the model predicts. Clear separation prevents leakage and ensures consistent training and inference behavior.
Category: Misunderstanding
What are common Feature types?
Numeric, categorical, ordinal, text embeddings, image features, time-based aggregations, and interaction terms. Each type requires appropriate preprocessing and validation.
Category: Overview
How are Features maintained in production?
Use feature stores, versioning, backfills, and monitoring for drift and data quality. Enforce contracts and compute features consistently for training and serving.
Category: Management
What tools help create Features?
Libraries like scikit-learn, Spark, and dbt for transformations; feature stores like Feast streamline reuse and online/offline consistency.
Category: Platform
How do Features relate to model fairness?
Feature choices can encode bias. Auditing, fairness constraints, and causal reasoning help ensure features do not proxy protected attributes.
Category: Risk