Argument-based Explanation Functions

Explaining predictions made by inductive classifiers whose internal reasoning is left unspecified (black-boxes) has become an important topic. Abductive explanations are one of the most popular types of explanations that are provided for the purpose. They are sufficient reasons for making predictions. They are generated from the whole feature space, which is not reasonable in practice. This paper investigates functions that generate abductive explanations from a set of instances. It shows that such explainers should be defined with great care since they cannot satisfy two desirable properties at the same time, namely existence of explanations for every individual decision (success) and correctness of explanations (coherence). The paper provides a general argumentation-based setting in which various functions satisfying one of the two properties are defined.