Analyzing Error Sources in Global Feature Effect Estimation
Timo Heiß, Coco Bögel, Bernd Bischl, Giuseppe Casalicchio · 2026
Read Analyzing Error Sources in Global Feature Effect Estimation by Timo Heiß, Coco Bögel, Bernd Bischl, Giuseppe Casalicchio (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-2603-15057.txt.
Description
Heiss and co-authors break down the error sources behind global feature effect plots, such as partial dependence and accumulated local effects. These plots are estimates, and their trust level was poorly understood. The team splits mean squared error into model bias, estimation bias, and two variance parts. Simulations across many data processes and learners test training, holdout, and cross-validation strategies. Holdout data is cleanest in theory, but training-data bias proved tiny in practice. Cross-validation cut model variance, notably for overfit models.
Catalog record
| CREATOR | Timo Heiß, Coco Bögel, Bernd Bischl, Giuseppe Casalicchio |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 78506 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2603.15057