Axiomatic Characterization of Data-Driven Influence Measures for Classification
We study the following problem: given a labeled dataset and a specific datapoint x, how did the i-th feature influence the classification for x? We identify a family of numerical influence measures - functions that, given a datapoint x, assign a numeric value phi_i(x) to every feature i, corresponding to how altering i's value would influence the outcome for x. This family, which we term monotone influence measures (MIM), is uniquely derived from a set of desirable properties, or axioms. The MIM family constitutes a provably sound methodology for measuring feature influence in classification domains; the values generated by MIM are based on the dataset alone, and do not make any queries to the classifier. While this requirement naturally limits the scope of our framework, we demonstrate its effectiveness on data.
Code (0)
등록된 구현이 없습니다.
Tasks
ClassificationGeneral ClassificationSimilar Papers 제목 키워드 기반
Model Explanations via the Axiomatic Causal Lens
Explaining the decisions of black-box models is a central theme in the study of trustworthy ML. Numerous measures have been proposed in the literature; however, none of them take an axiomatic approach to causal explainab…
Feature ImportancemodelHow to choose the most appropriate centrality measure? A decision tree approach
Centrality metrics play a crucial role in network analysis, while the choice of specific measures significantly influences the accuracy of conclusions as each measure represents a unique concept of node importance. Among…
On evaluation of joint risk for non-negative multivariate risks under dependence uncertainty
In this paper, we propose a novel axiomatic approach to evaluating the joint risk of multiple insurance risks under dependence uncertainty. Motivated by both the theory of expected utility and the Cobb-Dauglas utility fu…
On the influence of dependent features in classification problems: a game-theoretic perspective
This paper deals with a new measure of the influence of each feature on the response variable in classification problems, accounting for potential dependencies among certain feature subsets. Within this framework, we con…
Elicitability of Return Risk Measures
Informally, a risk measure is said to be elicitable if there exists a suitable scoring function such that minimizing its expected value recovers the risk measure. In this paper, we analyze the elicitability properties of…