Unjustified Classification Regions and Counterfactual Explanations in Machine Learning

A Survey of Contrastive and Counterfactual Explanation Generation Methods for Explainable Artificial Intelligence

“…The candidate counterfactuals are then clustered, as the initial local neighbourhood is updated to become a more extensive hyperspherical layer, until it can no longer be extended. Laugel et al [100] enhance the work on justified counterfactual explanations. They argue that the distance from the test instance to a counterfactual does not sufficiently measure counterfactual's relevance, as the counterfactual in question may appear disconnected from the ground-truth data.…”

“…interpretable intermediate predictors) mimic the local neighbourhood (i.e., fidelity) and the data example to be explained (i.e., hit). Laugel et al measure how justified counterfactuals are by averaging a binary score (one if the explanation is justified following the proposed definition, zero otherwise) over all the generated explanations [100], [144]. It is worth noting that the run-time of explanation generation algorithms is reported in addition to the evaluation metrics for several frameworks [132], [139], [146], [152], [156], [159].…”

“…Fernández [91] (where X is the data example requiring an explanation, A is the class predicted for X , B is the contrast-class in question, C and D are specific visual patterns present or absent in the given video frame X ). On a similar note, Laugel et al treat a counterfactual explanation as a specific data instance, close to the observation whose prediction is explained, but predicted to belong to a different class [100]. Kostic defines a counterfactual to be a statement describing a hypothetically different situation to the actual state of affairs [96].…”

“…Indeed, contfactuals are particularly suitable for informing the end-user why a given data example is assigned a particular class label. 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], [141], [145], [148], [151], or combinations of these [100], [105], [134], [152], [154], [160]. In three studies [67], [128], [137], the classifiers used in the experiments are not specified.…”

“…Finally, there is a number of marginal interpretations of contfactuals among the RQ 3 -related studies. Laugel et al denote a counterfactual as a specific data instance that changes the algorithm's prediction [100]. Poyiadzi et al denote a counterfactual to be ''the new state of the object'' [49].…”