Interventional SHAP Values and Interaction Values for Piecewise Linear Regression Trees

In recent years, game-theoretic Shapley values have gained increasing attention with respect to local model explanation by feature attributions. While the approach using is model-independent, their (exact) computation usually intractable, so efficient model-specific algorithms been devised including approaches for decision trees or ensembles in general. Our work goes further this direction extending interventional TreeSHAP algorithm piecewise linear regression trees, which more past few years. To end, we introduce a decomposition of contribution function based on paths, allows comprehensible formulation SHAP tree-based models. can also be readily applied computing interaction these particular, as main paper, provide models precomputing statistics background data tree structure.