Table 5.

Modelling stage and different technical solutions for algorithm explainability and interpretability.

stage/method	technical solution
in-processing/model-specific	rule-based explanations: decision trees, rule-induction methods model’s coefficients: linear regression, linear discriminant analysis nearest prototype: k-nearest-neighbour, naive Bayes
post-processing/model-agnostic	surrogate explanations: LIME [90], explainable boosting machines [91], PIRL [92] perturbation: gradient-based attribution methods [93], permutation Importance [94], SHAP [95] simulation analysis (what-if?): counterfactual explanations and algorithmic recourse [96,97]