On the Robustness of Global Feature Effect Explanations
This addresses the robustness issue in model debugging and scientific discovery for applied sciences, but it is incremental as it builds on existing explanation methods.
The paper tackles the vulnerability of global post-hoc explanations like partial dependence plots and accumulated local effects to data and model perturbations in black-box supervised learning, introducing theoretical bounds and experimental results that quantify the gap between best and worst-case scenarios for interpreting predictions.
We study the robustness of global post-hoc explanations for predictive models trained on tabular data. Effects of predictor features in black-box supervised learning are an essential diagnostic tool for model debugging and scientific discovery in applied sciences. However, how vulnerable they are to data and model perturbations remains an open research question. We introduce several theoretical bounds for evaluating the robustness of partial dependence plots and accumulated local effects. Our experimental results with synthetic and real-world datasets quantify the gap between the best and worst-case scenarios of (mis)interpreting machine learning predictions globally.