cc-Shapley: Measuring Multivariate Feature Importance Needs Causal Context
Jörg Martin, Stefan Haufe · 2026
Read cc-Shapley: Measuring Multivariate Feature Importance Needs Causal Context by Jörg Martin, Stefan Haufe (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-2602-20396.txt.
Description
Jörg Martin and Stefan Haufe argue that multivariate feature importance needs causal context. Explainable AI aims to reveal which features matter, letting people check models or even spark discoveries. Yet the standard Shapley-value approach, used in a purely data-driven way, can mislead. Even with two features, collider bias and suppression can create spurious links when one feature is judged only in the observational context of the other.
Causal knowledge of the data-generating process is needed to spot and correct such misleading attributions. The authors propose cc-Shapley, an interventional version of standard Shapley values that uses the causal structure to judge each feature in the causal context of the rest. Theory shows it removes spurious links from collider bias. Tests on synthetic and real-world datasets show links vanishing or reversing when moving from observational to cc-Shapley values, against univariate importance.
Catalog record
| CREATOR | Jörg Martin, Stefan Haufe |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 72 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2602-20396