Conformal outlier detection wrapper (v3.3 backlog item 3)
Author: JayeshSuryavanshiCreated Sep 13, 2026Updated Sep 13, 2026
Hi, I noticed item 3 in the v3.3 backlog is a conformal wrapper that works with any BaseDetector. I'd like to work on it if nobody else has picked it up.
What I have in mind is a ConformalDetector that fits the base detector, keeps a calibration split aside, converts scores to conformal p-values (Bates et al., "Testing for outliers with conformal p-values", Annals of Statistics 2023) and applies Benjamini-Hochberg, so you can set a false discovery rate instead of a contamination fraction. Only numpy/scipy, no new dependency.
No rush on my side. I'd rather wait until #716 and #717 are reviewed before sending the PR. Would that be okay?
Source: yzhao062/pyod