Scoring & evidence
Confidence interval
A range expressing uncertainty around an estimated score.
Definition
A confidence interval estimates the range within which the underlying performance measure is likely to fall given sampling uncertainty.
Why it matters
Small score differences may not be meaningful when intervals overlap or the test set is small.
Related concepts
- Metric
A defined quantity used to assess an aspect of performance.
- Leaderboard
A ranked table of systems evaluated under stated conditions.