![[2508.09639] UbiQTree: Uncertainty Quantification in XAI with Tree Ensembles](https://arxiv.org/static/browse/0.3.4/images/arxiv-logo-fb.png){kind=logo}
[2508.09639] UbiQTree: Uncertainty Quantification in XAI with Tree Ensembles
Summary
The paper presents UbiQTree, a method for decomposing uncertainty in SHAP values used in explainable AI, focusing on aleatoric and epistemic uncertainties in tree ensembles.
Why It Matters
Understanding uncertainty in AI models is crucial, especially in high-stakes fields like healthcare. This research enhances the interpretability and reliability of SHAP values, guiding better decision-making and model refinement.
Key Takeaways
- UbiQTree decomposes SHAP value uncertainty into aleatoric and epistemic components.
- The method integrates Dempster-Shafer theory and Dirichlet processes for improved uncertainty quantification.
- Findings indicate that high SHAP values may not correlate with stable features.
- Better data representation can reduce epistemic uncertainty in model predictions.
- Tree-based models, particularly bagging, are effective for quantifying uncertainty.
Computer Science > Artificial Intelligence arXiv:2508.09639 (cs) [Submitted on 13 Aug 2025 (v1), last revised 24 Feb 2026 (this version, v2)] Title:UbiQTree: Uncertainty Quantification in XAI with Tree Ensembles Authors:Akshat Dubey, Aleksandar Anžel, Bahar İlgen, Georges Hattab View a PDF of the paper titled UbiQTree: Uncertainty Quantification in XAI with Tree Ensembles, by Akshat Dubey and 3 other authors View PDF HTML (experimental) Abstract:Explainable Artificial Intelligence (XAI) techniques, such as SHapley Additive exPlanations (SHAP), have become essential tools for interpreting complex ensemble tree-based models, especially in high-stakes domains such as healthcare analytics. However, SHAP values are usually treated as point estimates, which disregards the inherent and ubiquitous uncertainty in predictive models and data. This uncertainty has two primary sources: aleatoric and epistemic. The aleatoric uncertainty, which reflects the irreducible noise in the data. The epistemic uncertainty, which arises from a lack of data. In this work, we propose an approach for decomposing uncertainty in SHAP values into aleatoric, epistemic, and entanglement components. This approach integrates Dempster-Shafer evidence theory and hypothesis sampling via Dirichlet processes over tree ensembles. We validate the method across three real-world use cases with descriptive statistical analyses that provide insight into the nature of epistemic uncertainty embedded in SHAP explanations...