Skip to content

Real-Time Feature Computation

An approach in which feature values are computed close to prediction time instead of being fully precomputed.

Real-time feature computation becomes important in systems where behavioral or risk signals change rapidly. Pre-materialized features may not be fresh enough, so some values are computed online. This approach improves freshness, but also increases latency and infrastructural complexity. It therefore requires a careful balance between speed, recency, and cost.