On Procrustes Contamination in Machine Learning Applications of Geometric Morphometrics

Lloyd Austin Courtenay · 2026

Read On Procrustes Contamination in Machine Learning Applications of Geometric Morphometrics by Lloyd Austin Courtenay (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2601-18448.txt.

Description

Lloyd Austin Courtenay warns of a leak that can spoil studies which sort bones and shells by shape. Shape work first lines up all forms with a shared fit, so that size and tilt do not skew the chart. Teams then teach a model to sort the lined up forms into groups. The snag is that the shared fit is done on all forms at once, before the split into train and test sets.

That early blend lets facts from the test set seep into the train set. A model can then look sharp in tests for the wrong cause. Courtenay sets out how the leak works and why it counts for shape based claims. The fix is to keep the fit and the split apart, so each test stays fair. Shape fields that now lean on smart models need that care to keep their finds sound.

Catalog record

CREATORLloyd Austin Courtenay
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE45 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2601-18448

Browse the full library · About the Tower of Babel