What is a feature store in classification systems?

A feature store is a centralized system for managing, storing, and serving features consistently for training and inference.

In large-scale classification systems, feature inconsistency between training and inference is a major problem. A feature store solves this by providing a single source of truth for feature computation. It ensures features are computed once and reused across offline training pipelines and real-time serving systems. It typically includes offline storage (data lake/warehouse), online storage (low-latency DB), and a feature registry. It also supports feature versioning, lineage tracking, and point-in-time correctness to prevent data leakage.