ExplainabilityTransparencyAccountabilitycs.LG
Beyond TreeSHAP: Efficient Computation of Any-Order Shapley Interactions for Tree Ensembles
Maximilian Muschalik, Fabian Fumagalli, Barbara Hammer, Eyke Hüllermeier
January 22, 2024
AAAI Conference on Artificial Intelligence
50 citations
2 influential
Key Takeaways
-
Tree ensemble models dominate tabular data but lack transparency: Gradient-boosted trees and other ensemble models achieve state-of-the-art performance on many real-world problems involving structured data, yet they remain "black box" models that are difficult to interpret—TreeSHAP-IQ addresses this interpretability gap.
-
Understanding feature interactions matters for responsible AI: Beyond explaining individual feature importance, this research enables analysis of how multiple features interact to influence predictions, which is crucial for identifying potential biases, unexpected model behaviors, or unreliable decision patterns in high-stakes applications.
-
Computational efficiency enables practical explainability: TreeSHAP-IQ computes interaction explanations in a single pass through the model using polynomial arithmetic, making it computationally feasible to explain complex ensemble predictions without prohibitive computational costs that would limit real-world deployment.
-
Shapley values provide mathematically principled explanations: The research builds on Shapley values, a game-theoretic concept that offers a theoretically sound and fair method for attributing model predictions to features, rather than relying on ad-hoc explanation techniques that may be misleading.
Abstract
While shallow decision trees may be interpretable, larger ensemble models like gradient-boosted trees, which often set the state of the art in machine learning problems involving tabular data, still remain black box models. As a remedy, the Shapley value (SV) is a well-known concept in explainable artificial intelligence (XAI) research for quantifying additive feature attributions of predictions. The model-specific TreeSHAP methodology solves the exponential complexity for retrieving exact SVs from tree-based models. Expanding beyond individual feature attribution, Shapley interactions reveal the impact of intricate feature interactions of any order. In this work, we present TreeSHAP-IQ, an efficient method to compute any-order additive Shapley interactions for predictions of tree-based models. TreeSHAP-IQ is supported by a mathematical framework that exploits polynomial arithmetic to compute the interaction scores in a single recursive traversal of the tree, akin to Linear TreeSHAP. We apply TreeSHAP-IQ on state-of-the-art tree ensembles and explore interactions on well-established benchmark datasets.