Skip to content
ML Academy

Introduced on day 27

Forecast headroom

The gap between the best simple baseline's error and the best error any model could achieve given irreducible noise; the most a heavier model can gain.

In detail

On real data the floor is not observed, so headroom is estimated from signal-to-noise, from how much each added baseline has already recovered and from segment-level diagnostics. In a simulation with a known true mean the floor can be measured directly, which makes headroom a teaching device and an upper bound.