A Counterfactual Theory of Causal Explanation

The outlined classification-oriented frameworks are evaluated on classifiers based on logistic regression, decision trees, gradient boosted decision trees, support vector machines, random forests, neural networks, or combinations of these.

“Thus, the outlined classification-oriented frameworks are evaluated on classifiers based on logistic regression [55], [136], [153], [158], decision trees [46], [80], [122], [140], [150], [155], [159], gradient boosted decision trees [147], support vector machines [131], [138], [146], random forests [81], [86], [142]-[144], neural networks [6], [48], [49], [91], [129], [130], [133], [135], [139], [”

In order to guarantee the coherence of the counterfactual data example used for explanation, Russell et al. propose an integer programming-based method [153].

“In order to guarantee the coherence of the counterfactual data example used for explanation, an integer programming-based method is proposed.”

Russell et al.'s framework [153] is limited to: (1) explaining predictions of only linear classifiers and (2) a simple structure of the textual explanation template.

“However, this framework is limited to: (1) explaining predictions of only linear classifiers and (2) a simple structure of the textual explanation template.”

Russell et al. claim that the generated counterfactual explanations are diverse, as diversity constraints are applied iteratively to a set of candidate counterfactuals [153].

“In addition, the generated counterfactual explanations are claimed to be diverse, as diversity constraints are applied iteratively to a set of candidate counterfactuals.”

Russell et al. adopt a similar approach to encompass continuous and discrete variables as well as the combination of the two [153].

“Russell et al. adopt a similar approach to encompass continuous and discrete variables as well as the combination of the two [153].”

The main focus of Russell et al.'s work [153] is placed on assessing coherence and diversity of generated counterfactuals.

“The main focus of the work is however placed on assessing coherence and diversity of generated counterfactuals.”

Multiple-sentence counterfactual explanations provide end-users with specific details for the given decision or explain multiple decisions at once.

“In contrast, multiple-sentence explanations provide end-users with specific details for the given decision [153], [156], [157], [161] or explain multiple decisions at once [146].”

Some publications stress that human evaluation should be encouraged to estimate the quality and effectiveness of the generated counterfactuals.

“However, whereas certain publications do not provide any specific evaluation method, some do stress that human evaluation should be encouraged to estimate the quality and effectiveness of the generated counterfactuals [46], [150], [153].”