AcME -- Accelerated Model-agnostic Explanations: Fast Whitening of the Machine-Learning Black Box

• DOI: 10.48550/arxiv.2112.12635
• Publication Date: 2021-01-01
• Authors: David Dandolo; Chiara Masiero; Mattia Carletti; Davide Dalle Pezze; Gian Antonio Susto
• Keywords: Interpretability; Feature (linguistics); Black box

In the context of human-in-the-loop Machine Learning applications, like Decision Support Systems, interpretability approaches should provide actionable insights without making users wait. this paper, we propose Accelerated Model-agnostic Explanations (AcME), an approach that quickly provides feature importance scores both at global and local level. AcME can be applied a posteriori to each regression or classification model. Not only does compute ranking, but it also what-if analysis tool assess how changes in features values would affect model predictions. We evaluated proposed on synthetic real-world datasets, comparison with SHapley Additive exPlanations (SHAP), drew inspiration from, which is currently one state-of-the-art model-agnostic approaches. achieved comparable results terms quality produced explanations while reducing dramatically computational time providing consistent visualization for interpretations. To foster research field, sake reproducibility, repository code used experiments.