TheOut-of-DistributionProbleminExplainability andSearchMethodsforFeatureImportance Explanations

To resolvethis issue, we propose asimple alteration to the model training process, which results in more socially aligned explanations and metrics.