dc.contributor.author
Dubey, Akshat
dc.contributor.author
Anžel, Aleksandar
dc.contributor.author
İlgen, Bahar
dc.contributor.author
Hattab, Georges
dc.date.accessioned
2026-06-09T07:52:44Z
dc.date.available
2026-06-09T07:52:44Z
dc.identifier.uri
https://refubium.fu-berlin.de/handle/fub188/52690
dc.identifier.uri
http://dx.doi.org/10.17169/refubium-52420
dc.description.abstract
Explainable artificial intelligence (XAI) techniques, particularly Shapley additive explanations (SHAP), are essential for interpreting ensemble tree-based models in critical areas such as healthcare. However, SHAP values are often treated as point estimates that neglect uncertainty originating from aleatoric (irreducible noise) and epistemic (lack of data) sources. This work introduces an approach that decomposes SHAP value uncertainty into aleatoric, epistemic, and entanglement components. This approach employs Dempster-Shafer evidence theory and Dirichlet process (DP) hypothesis sampling over tree ensembles. The use-case validation reveals insights into epistemic uncertainty within SHAP explanations, enhancing the reliability and interpretability of SHAP attributions. This informs robust decision-making and model refinement. Our findings suggest that reducing epistemic uncertainty requires improved data quality and model development techniques. Tree-based models, particularly bagging, are effective in quantifying such uncertainties.
en
dc.format.extent
26 Seiten
dc.rights.uri
https://creativecommons.org/licenses/by/4.0/
dc.subject
machine learning
en
dc.subject
explainability
en
dc.subject
random forest
en
dc.subject
ensemble machine learning
en
dc.subject.ddc
000 Informatik, Informationswissenschaft, allgemeine Werke::000 Informatik, Wissen, Systeme::004 Datenverarbeitung; Informatik
dc.title
UbiQTree: Uncertainty quantification in XAI with tree ensembles
dc.type
Wissenschaftlicher Artikel
dcterms.bibliographicCitation.articlenumber
101454
dcterms.bibliographicCitation.doi
10.1016/j.patter.2025.101454
dcterms.bibliographicCitation.journaltitle
Patterns
dcterms.bibliographicCitation.number
4
dcterms.bibliographicCitation.volume
7
dcterms.bibliographicCitation.url
https://doi.org/10.1016/j.patter.2025.101454
refubium.affiliation
Mathematik und Informatik
refubium.affiliation.other
Institut für Mathematik

refubium.resourceType.isindependentpub
no
dcterms.accessRights.openaire
open access
dcterms.isPartOf.eissn
2666-3899
refubium.resourceType.provider
WoS-Alert