Tree-based machine learning models such as random forests, decision trees and gradient boosted trees are popular nonlinear predictive models, yet comparatively little attention has been paid to explaining their predictions. Here we improve the interpretability of tree-based models through three main contributions. (1) A 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. We apply these tools to three medical machine learning problems and show how combining many high-quality local explanations allows us to represent global structure while retaining local faithfulness to the original model. These tools enable us to (1) identify high-magnitude but low-frequency nonlinear mortality risk factors in the US population, (2) highlight distinct population subgroups with shared risk characteristics, (3) identify nonlinear interaction effects among risk factors for chronic kidney disease and (4) 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.
Subscribe to Journal
Get full journal access for 1 year
only $8.67 per issue
All prices are NET prices.
VAT will be added later in the checkout.
Rent or Buy article
Get time limited or full article access on ReadCube.
All prices are NET prices.
The pre-processed mortality data are available at http://github.com/suinleelab/treexplainer-study. Privacy restrictions prevent the release of the hospital procedure-related data, and the kidney disease data are only available directly from the National Institute of Diabetes, Digestive and Kidney Diseases (NIDDK).
Code supporting this paper is published online at https://github.com/suinleelab/treexplainer-study. A widely used Python implementation of TreeExplainer is available at https://github.com/slundberg/shap, and portions of it are included in the standard release of XGBoost (https://xgboost.ai), LightGBM (https://github.com/Microsoft/LightGBM) and CatBoost (https://catboost.ai).
The state of data science & maching learning. Kaggle https://www.kaggle.com/surveys/2017 (2017).
Friedman, J., Hastie, T. & Tibshirani, R. The Elements of Statistical Learning Vol. 1 (Springer Series in Statistics, Springer, 2001).
Lundberg, S. M. & Lee, S.-I. A unified approach to interpreting model predictions. Adv. Neural Inf. Process. Syst. 30, 4768–4777 (2017).
Saabas, A. treeinterpreter python package. GitHub https://github.com/andosa/treeinterpreter (2019).
Ribeiro, M. T., Singh, S. & Guestrin, C. Why should i trust you?: Explaining the predictions of any classifier. In Proc. 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining 1135–1144 (ACM, 2016).
Datta, A., Sen, S. & Zick, Y. Algorithmic transparency via quantitative input influence: theory and experiments with learning systems. In Proc. 2016 IEEE Symposium on Security and Privacy (SP), 598–617 (IEEE, 2016).
Štrumbelj, E. & Kononenko, I. Explaining prediction models and individual predictions with feature contributions. Knowl. Inf. Syst. 41, 647–665 (2014).
Baehrens, D. et al. How to explain individual classification decisions. J. Mach. Learn. Res. 11, 1803–1831 (2010).
Shapley, L. S. A value for n-person games. Contrib. Theor. Games 2, 307–317 (1953).
Sundararajan, M. & Najmi, A. The many Shapley values for model explanation. Preprint at https://arxiv.org/abs/1908.08474 (2019).
Janzing, D., Minorics, L. & Blöbaum, P. Feature relevance quantification in explainable AI: a causality problem. Preprint at https://arxiv.org/abs/1910.13413 (2019).
Matsui, Y. & Matsui, T. NP-completeness for calculating power indices of weighted majority games. Theor. Comput. Sci. 263, 305–310 (2001).
Fujimoto, K., Kojadinovic, I. & Marichal, J.-L. Axiomatic characterizations of probabilistic and cardinal-probabilistic interaction indices. Games Econ. Behav. 55, 72–99 (2006).
Ribeiro, M. T., Singh, S. & Guestrin, C. Anchors: high-precision model-agnostic explanations. In Proc. AAAI Conference on Artificial Intelligence (2018).
Shortliffe, E. H. & Sepúlveda, M. J. Clinical decision support in the era of artificial intelligence. JAMA 320, 2199–2200 (2018).
Lundberg, S. M. et al. Explainable machine learning predictions to help anesthesiologists prevent hypoxemia during surgery. Nat. Biomed. Eng. 2, 749–760 (2018).
Cox, C. S. et al. Plan and operation of the NHANES I Epidemiologic Followup Study, 1992. Vital Health Stat. 35, 1–231 (1997).
Chen, T. & Guestrin, C. Xgboost: a scalable tree boosting system. In Proc. 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining 785–794 (ACM, 2016).
Haufe, S. et al. On the interpretation of weight vectors of linear models in multivariate neuroimaging. Neuroimage 87, 96–110 (2014).
Kim, B. et al. Interpretability beyond feature attribution: quantitative testing with concept activation vectors (TCAV). In International Conference on Machine Learning (ICLR, 2018).
Yosinski, J., Clune, J., Nguyen, A., Fuchs, T. & Lipson, H. Understanding neural networks through deep visualization. In ICML Deep Learning Workshop (ICML, 2015).
Bau, D., Zhou, B., Khosla, A., Oliva, A. & Torralba, A. Network dissection: quantifying interpretability of deep visual representations. In Proc. IEEE Conference on Computer Vision and Pattern Recognition 6541–6549 (IEEE, 2017).
Leino, K., Sen, S., Datta, A., Fredrikson, M. & Li, L. Influence-directed explanations for deep convolutional networks. In Proc. 2018 IEEE International Test Conference (ITC) 1–8 (IEEE, 2018).
Group, S. R. A randomized trial of intensive versus standard blood-pressure control. N. Engl. J. Med. 373, 2103–2116 (2015).
Mozaffarian, D. et al. Heart disease and stroke statistics-2016 update a report from the American Heart Association. Circulation 133, e38–e48 (2016).
Bowe, B., Xie, Y., Xian, H., Li, T. & Al-Aly, Z. Association between monocyte count and risk of incident CKD and progression to ESRD. Clin. J. Am. Soc. Nephrol. 12, 603–613 (2017).
Fan, F., Jia, J., Li, J., Huo, Y. & Zhang, Y. White blood cell count predicts the odds of kidney function decline in a Chinese community-based population. BMC Nephrol. 18, 190 (2017).
Zinkevich, M. Rules of machine learning: best practices for ML engineering (2017).
van Rooden, S. M. et al. The identification of Parkinson’s disease subtypes using cluster analysis: a systematic review. Mov. Disord. 25, 969–978 (2010).
Sørlie, T. et al. Repeated observation of breast tumor subtypes in independent gene expression data sets. Proc. Natl Acad. Sci. USA 100, 8418–8423 (2003).
Lapuschkin, S. et al. Unmasking clever hans predictors and assessing what machines really learn. Nat. Commun. 10, 1096 (2019).
Pfungst, O. Clever Hans: (the Horse of Mr. Von Osten.) A Contribution to Experimental Animal and Human Psychology (Holt, Rinehart and Winston, 1911).
Machine Learning Recommendations for Policymakers (IIF, 2019); https://www.iif.com/Publications/ID/3574/Machine-Learning-Recommendations-for-Policymakers
Deeks, A. The judicial demand for explainable artificial intelligence. (2019).
Plumb, G., Molitor, D. & Talwalkar, A. S. Model agnostic supervised local explanations. Adv. Neural Inf. Process. Syst. 31, 2520–2529 (2018).
Young, H. P. Monotonic solutions of cooperative games. Int. J. Game Theor. 14, 65–72 (1985).
Ancona, M., Ceolini, E., Oztireli, C. & Gross, M. Towards better understanding of gradient-based attribution methods for deep neural networks. In Proc. 6th International Conference on Learning Representations (ICLR 2018) (2018).
Hooker, S., Erhan, D., Kindermans, P.-J. & Kim, B. A benchmark for interpretability methods in deep neural networks. In Conference on Neural Information Processing Systems (NIPS, 2019).
Shrikumar, A., Greenside, P., Shcherbina, A. & Kundaje, A. Not just a black box: learning important features through propagating activation differences. Preprint at https://arxiv.org/abs/1605.01713 (2016).
Lunetta, K. L., Hayward, L. B., Segal, J. & Van Eerdewegh, P. Screening large-scale association study data: exploiting interactions using random forests. BMC Genet. 5, 32 (2004).
Jiang, R., Tang, W., Wu, X. & Fu, W. A random forest approach to the detection of epistatic interactions in case-control studies. BMC Bioinformatics 10, S65 (2009).
We are grateful to R. Chen, A. Okeson, C. Robinson, V. Khotilovich, N. Hiranuma, J. Janizek, M. T. Ribeiro, J. Schreiber, P. Hall and members of S.-I.L.’s group for the feedback and assistance they provided during the development and preparation of this research. This work was funded by the National Science Foundation (DBI-1759487, DBI-1552309, DBI-1355899, DGE-1762114 and DGE-1256082), American Cancer Society (127332-RSG-15-097-01-TBG), National Institutes of Health (R35 GM 128638 and R01 NIA AG 061132), and an unrestricted gift from the Northwest Kidney Centers to the University of Washington Kidney Research Institute. The Chronic Renal Insufficiency Cohort (CRIC) study was conducted by the CRIC investigators and supported by the National Institute of Diabetes and Digestive and Kidney Diseases (NIDDK). The data from the CRIC study reported here were supplied by the NIDDK Central Repositories. This manuscript was not prepared in collaboration with Investigators of the CRIC study and does not necessarily reflect the opinions or views of the CRIC study, the NIDDK Central Repositories or the NIDDK.
The authors declare no competing interests.
Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
About this article
Cite this article
Lundberg, S.M., Erion, G., Chen, H. et al. From local explanations to global understanding with explainable AI for trees. Nat Mach Intell 2, 56–67 (2020). https://doi.org/10.1038/s42256-019-0138-9