Model-agnostic variable importance for predictive uncertainty: an entropy-based approach
This work addresses the need for interpretability in probabilistic models to build trust, though it is incremental as it adapts existing methods rather than introducing a new paradigm.
The paper tackles the problem of explaining predictive uncertainty in machine learning models by extending existing explainability methods to uncertainty-aware models, demonstrating that these adaptations can measure feature impacts on entropy and log-likelihood with experiments on synthetic and real-world data.
In order to trust the predictions of a machine learning algorithm, it is necessary to understand the factors that contribute to those predictions. In the case of probabilistic and uncertainty-aware models, it is necessary to understand not only the reasons for the predictions themselves, but also the reasons for the model's level of confidence in those predictions. In this paper, we show how existing methods in explainability can be extended to uncertainty-aware models and how such extensions can be used to understand the sources of uncertainty in a model's predictive distribution. In particular, by adapting permutation feature importance, partial dependence plots, and individual conditional expectation plots, we demonstrate that novel insights into model behaviour may be obtained and that these methods can be used to measure the impact of features on both the entropy of the predictive distribution and the log-likelihood of the ground truth labels under that distribution. With experiments using both synthetic and real-world data, we demonstrate the utility of these approaches to understand both the sources of uncertainty and their impact on model performance.