Towards Explainability of Tree-Based Ensemble Models. A Critical Overview

Tree-based ensemble models are widely applied in artificial intelligence systems due to their robustness and generality. However, those models are not transparent. For the sake of making systems trustworthy and dependable, multiple explanation techniques are developed. This paper presents selected explainability techniques for tree-based ensemble models. First, the aspect of black-boxness and the definition of explainability are reported. Then, predominant model-agnostic (LIME, SHAP, counterfactual explanations), as well as model-specific techniques (fusion into a single decision tree, iForest) are described. Moreover, other methods are also briefly mentioned. Finally, a brief summary is presented.