Explications contrefactuelles pour les forêts aléatoires prudentes


Cautious random forests are designed to make indeterminate decisions when the uncertainty is too high. Since indeterminacy has a cost, it seems desirable to highlight why a precise decision could not be made for an instance, or which minimal modifications can be made to the instance so that the decision becomes a single class. In this paper, we propose to use the notion of counterfactual and propose an efficient algorithm to generate determinate counterfactual examples for cautious random forests. We evaluate the efficiency of our strategy on different datasets and we illustrate its utility on two simple case studies involving both tabular and image data.