New method enhances conditional Shapley feature importance inference

Researchers have developed a new method for inferring conditional Shapley feature importance, addressing limitations in existing estimators that often provide only point quantities and fail to account for feature dependency. The proposed approach integrates out out-of-coalition features under their true conditional distribution, aiming for a global, loss-based importance measure. This method is theoretically proven to be $\sqrt{n}$-consistent and asymptotically normal, with empirical studies showing accurate confidence interval coverage and Type-I error control. AI

IMPACT Enhances interpretability of machine learning models by improving feature attribution methods.

RANK_REASON The item is an academic paper detailing a new statistical methodology. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv stat.ML →

• Bonferroni
• K-fold cross-fitting
• Sage
• Shapley Values
• UCI Concrete
• U-statistic

• paper
• other

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