On Correcting SHAP Scores

arXiv:2405.00076v1 Announce Type: new
Abstract: Recent work uncovered examples of classifiers for which SHAP scores yield misleading feature attributions. While such examples might be perceived as suggesting the inadequacy of Shapley values for explainability, this paper shows that the source of the identified shortcomings of SHAP scores resides elsewhere. Concretely, the paper makes the case that the failings of SHAP scores result from the characteristic functions used in earlier works. Furthermore, the paper identifies a number of properties that characteristic …

abstract arxiv case classifiers cs.ai cs.lg examples explainability feature paper shap shows type values while work