Mutual information-based group explainers with coalition structure for machine learning model explanations

In this article, we propose and investigate ML group explainers in a general game-theoretic setting with the focus on coalitional game values and games based on the conditional and marginal expectation of an ML model. The conditional game takes into account the joint distribution of the predictors, while the marginal game depends on the structure of the model. The objective of the article is to unify the two points of view under predictor dependencies and to reduce the complexity of group explanations. To achieve this, we propose a feature grouping technique that employs an information-theoretic measure of dependence and design appropriate groups explainers. Furthermore, in the context of coalitional game values with a two-step formulation, we introduce a theoretical scheme that generates recursive coalitional game values under a partition tree structure and investigate the properties of the corresponding group explainers.