In a time where everybody is considering “deep learning” as “the only” machine learning method, it is pleasant to see that for transparent machine learning, when a “black-box” method may be problematic due to its critical application, alternative methods are highly relevant. On May, 11, 2019 the group of the Lee Lab (Explainable AI for Computational Biology & Medicine) released such a work:
Scott M. Lundberg, Gabriel Erion, Hugh Chen, Alex DeGrave, Jordan M. Prutkin, Bala Nair, Ronit Katz, Jonathan Himmelfarb, Nisha Bansal, Su-In Lee (2019): Explainable AI for Trees: From Local Explanations to Global Understanding. In: arXiV:1905.04610
Interestingly, tree-based machine learning models are among the “oldest” models in AI and are still the most popular non-linear predictive models used in practice today, but until recently ist was rare to use their power of explaining their predictions. In this work the authors improve the interpretability of tree-based models through 3 main contributions: 1) The first polynomial time algorithm to compute optimal explanations based on game theory. 2) A new type of explanation that directly measures local feature interaction effects. 3) A new set of tools for understanding global model structure based on combining many local explanations of each prediction. The authors apply these tools to 3 medical machine learning use cases and demonstrate how combining many high-quality local explanations allow to represent global structure while retaining local faithfulness to the original model. These tools are enabling to i) identify high magnitude but low frequency non-linear mortality risk factors , ii) highlight distinct population sub-groups with shared risk characteristics, iii) identify non-linear interaction effects among risk factors for chronic kidney disease, and iv) monitor a machine learning model deployed in a hospital by identifying which features are degrading the model’s performance over time. Given the popularity of tree-based machine learning models, these improvements to their interpretability have implications across a broad set of domains.
Read the original paper here: