Compute exact interval-valued attributions for tree ensembles

Develop exact computation methods for interval-valued attributions under both the pessimistic and averaging principles for tree ensembles, avoiding the potentially costly conversion of ensembles into single trees and addressing general virtual-sample budgets and large ensembles.

Background

The paper derives efficient exact procedures for single decision trees but only gives an approximate treatment for tree ensembles. An exact approach could convert an ensemble into a potentially very large single tree, although the resulting construction and leaf management may be computationally expensive. The authors also identify integer nonlinear programming and heuristic allocation methods as possible alternatives, while noting that general virtual-sample budgets and large ensembles remain unresolved computational cases.

References

As future work, exact computation under both principles could be extended to tree ensembles. One approach is to translate an ensemble into a potentially large single tree and apply the exact solutions presented in Section \ref{sec:leaf-perturbation}. However, constructing this tree and handling its many leaves may be costly. Alternatively, the optimal allocation problem could be formulated as an integer nonlinear program. The case s=1 may be handled efficiently using heuristic algorithms, whereas general values of s and large ensembles remain challenging.

— Interval-valued SHAP in Tree-Based Models  (2610.11953 - Zhu et al., 8 Oct 2026) in Section 6, Conclusion