Home / Curated / Algorithms / Tree SHAP

Tree SHAP

Algorithms AIMech:TreeSHAP DRAFT

A tree-specific algorithm computing Shapley feature attributions through recursive bookkeeping of weighted feature subsets along decision paths.

Identity Notes & Concept Scope This record denotes the original tree-traversal SHAP algorithm, not the TreeExplainer software package, all its later modes, or Independent Tree SHAP.

Claims & Grounded Evidence

2 Claims

Mechanism #explanation-method

The original procedure follows known-feature branches and uses node cover for missing-feature branches, while EXTEND and UNWIND operations track subset sizes and repeated split features to accumulate weighted leaf contributions.

supports Source: Consistent Individualized Feature Attribution for Tree Ensembles · Locator: Sections 3.1–3.2, Algorithms 1–2, PDF/printed pp. 3–4

Scope #subset-function-scope

The later method description explicitly defines the exact Tree SHAP calculation relative to the tree-traversal subset function and distinguishes Independent Tree SHAP.

supports Source: Explainable AI for Trees: From Local Explanations to Global Understanding · Locator: Methods 10.1–10.2, PDF/printed pp. 21–22

⚠️ Curated Limitations & Negative Results

Exact computation is relative to the specified tree-traversal subset-evaluation function; it is not an assertion of exact conditional expectations for arbitrary dependent input distributions.
Background/intervention variants and nonlinear output transforms require separate method assumptions.

Artificial Intelligence Ontology (AIO) Grounding

Mapped AIO Classes

AIO Target Class	Mapping Relation	Mapping Rationale
`AIO:DecisionTree` Decision Tree	related_match	The method directly explains tree-model predictions. The AIO target is described under classification, whereas the attribution method also applies to regression trees and ensembles; the link is related, not equivalence or a superclass assertion.

Evidence Sources & Primary Literature

2 Sources

Source ID	Publication / Document	Type / Access	Reference & Checksum
`original-paper`	Consistent Individualized Feature Attribution for Tree Ensembles Scott M. Lundberg, Gabriel G. Erion, Su-In Lee	primary paper full_text	ID: arXiv:1802.03888v3 sha256:4bb594a16893…
`tree-followup`	Explainable AI for Trees: From Local Explanations to Global Understanding Scott M. Lundberg, Gabriel Erion, Hugh Chen, Alex DeGrave, Jordan M. Prutkin, Bala Nair, Ronit Katz, Jonathan Himmelfarb, Nisha Bansal, Su-In Lee	primary paper full_text	ID: arXiv:1905.04610v1 sha256:aa3824ce44c7…

Curation Provenance

Curator: Codex

Method: ai_assisted

Model: GPT-6

Created: 2026-10-11T04:02:04Z

Updated: 2026-10-11T04:02:04Z

Source File: data/curated/TreeSHAP.yaml