Characterize coverage of tree-based Shapley values by imprecise Shapley values

Characterize the conditions under which imprecise Shapley values cover the Shapley values of tree-based models, thereby determining when imprecise Shapley values can serve as robustness indicators for tree-based explanations.

Background

The paper discusses imprecise SHAP, which was developed for imprecise-probability classifiers and produces interval-valued Shapley values reflecting uncertainty from insufficient training data. The authors note that the relationship between this framework and conventional Shapley values for tree-based models is not established. In particular, because the two approaches use different notions of Shapley values, it remains unresolved when the interval-valued imprecise values contain the corresponding tree-based Shapley values, a property that would be important for using them as robustness indicators.

References

While this is a generic approach with interesting theoretical properties, it is unclear how to leverage it to assess the robustness of Shapley values for tree-based models. In particular, due to differences in their notions of Shapley values, it is unclear under which conditions imprecise Shapley values can cover the Shapley values of tree-based models, which might be a crucial requirement for robustness indicators.

— Interval-valued SHAP in Tree-Based Models  (2610.11953 - Zhu et al., 8 Oct 2026) in Section 1, Introduction