New paper links SHAP model interpretation to functional ANOVA decomposition

A new paper published on arXiv explores the statistical underpinnings of SHAP (SHapley Additive exPlanations), a popular method for interpreting machine learning models. The research connects SHAP approximations to the functional ANOVA decomposition, highlighting how feature distribution choices and the number of estimated ANOVA terms impact SHAP's accuracy. While the connection to sensitivity analysis is insightful, the paper notes practical differences in constraints between machine learning explainability and traditional sensitivity analysis fields. AI

IMPACT Provides theoretical insights into model interpretability methods like SHAP, potentially influencing future research in explainable AI.

RANK_REASON Academic paper detailing statistical aspects of a machine learning methodology. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv stat.ML →

• alphaXiv
• Andrew Herren
• arXiv
• CatalyzeX
• DagsHub
• Functional Anova
• Gotit.pub
• Hugging Face
• ScienceCast
• Shap

• paper
• other

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →