Skip to main content

Thank you for visiting nature.com. You are using a browser version with limited support for CSS. To obtain the best experience, we recommend you use a more up to date browser (or turn off compatibility mode in Internet Explorer). In the meantime, to ensure continued support, we are displaying the site without styles and JavaScript.

From local explanations to global understanding with explainable AI for trees

Abstract

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.

A preprint version of the article is available at ArXiv.

Access options

Rent or Buy article

Get time limited or full article access on ReadCube.

from$8.99

All prices are NET prices.

Fig. 1: Local explanations based on TreeExplainer enable a wide variety of new ways to understand global model structure.
Fig. 2: Gradient boosted tree models can be more accurate than neural networks and more interpretable than linear models.
Fig. 3: Explanation method performance across 15 different evaluation metrics and 3 classification models in the chronic kidney disease dataset.
Fig. 4: By combining many local explanations, we can provide rich summaries of both an entire model and individual features.
Fig. 5: Monitoring plots reveal problems that would otherwise be invisible in a retrospective hospital machine learning model deployment.
Fig. 6: Local explanation embeddings support both supervised clustering and interpretable dimensionality reduction.

Data availability

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 availability

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).

References

  1. 1.

    The state of data science & maching learning. Kaggle https://www.kaggle.com/surveys/2017 (2017).

  2. 2.

    Friedman, J., Hastie, T. & Tibshirani, R. The Elements of Statistical Learning Vol. 1 (Springer Series in Statistics, Springer, 2001).

  3. 3.

    Lundberg, S. M. & Lee, S.-I. A unified approach to interpreting model predictions. Adv. Neural Inf. Process. Syst. 30, 4768–4777 (2017).

  4. 4.

    Saabas, A. treeinterpreter python package. GitHub https://github.com/andosa/treeinterpreter (2019).

  5. 5.

    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).

  6. 6.

    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).

  7. 7.

    Štrumbelj, E. & Kononenko, I. Explaining prediction models and individual predictions with feature contributions. Knowl. Inf. Syst. 41, 647–665 (2014).

    Article  Google Scholar 

  8. 8.

    Baehrens, D. et al. How to explain individual classification decisions. J. Mach. Learn. Res. 11, 1803–1831 (2010).

    MathSciNet  MATH  Google Scholar 

  9. 9.

    Shapley, L. S. A value for n-person games. Contrib. Theor. Games 2, 307–317 (1953).

    MathSciNet  MATH  Google Scholar 

  10. 10.

    Sundararajan, M. & Najmi, A. The many Shapley values for model explanation. Preprint at https://arxiv.org/abs/1908.08474 (2019).

  11. 11.

    Janzing, D., Minorics, L. & Blöbaum, P. Feature relevance quantification in explainable AI: a causality problem. Preprint at https://arxiv.org/abs/1910.13413 (2019).

  12. 12.

    Matsui, Y. & Matsui, T. NP-completeness for calculating power indices of weighted majority games. Theor. Comput. Sci. 263, 305–310 (2001).

    MathSciNet  Article  Google Scholar 

  13. 13.

    Fujimoto, K., Kojadinovic, I. & Marichal, J.-L. Axiomatic characterizations of probabilistic and cardinal-probabilistic interaction indices. Games Econ. Behav. 55, 72–99 (2006).

    MathSciNet  Article  Google Scholar 

  14. 14.

    Ribeiro, M. T., Singh, S. & Guestrin, C. Anchors: high-precision model-agnostic explanations. In Proc. AAAI Conference on Artificial Intelligence (2018).

  15. 15.

    Shortliffe, E. H. & Sepúlveda, M. J. Clinical decision support in the era of artificial intelligence. JAMA 320, 2199–2200 (2018).

    Article  Google Scholar 

  16. 16.

    Lundberg, S. M. et al. Explainable machine learning predictions to help anesthesiologists prevent hypoxemia during surgery. Nat. Biomed. Eng. 2, 749–760 (2018).

    Article  Google Scholar 

  17. 17.

    Cox, C. S. et al. Plan and operation of the NHANES I Epidemiologic Followup Study, 1992. Vital Health Stat. 35, 1–231 (1997).

    Google Scholar 

  18. 18.

    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).

  19. 19.

    Haufe, S. et al. On the interpretation of weight vectors of linear models in multivariate neuroimaging. Neuroimage 87, 96–110 (2014).

    Article  Google Scholar 

  20. 20.

    Kim, B. et al. Interpretability beyond feature attribution: quantitative testing with concept activation vectors (TCAV). In International Conference on Machine Learning (ICLR, 2018).

  21. 21.

    Yosinski, J., Clune, J., Nguyen, A., Fuchs, T. & Lipson, H. Understanding neural networks through deep visualization. In ICML Deep Learning Workshop (ICML, 2015).

  22. 22.

    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).

  23. 23.

    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).

  24. 24.

    Group, S. R. A randomized trial of intensive versus standard blood-pressure control. N. Engl. J. Med. 373, 2103–2116 (2015).

    Article  Google Scholar 

  25. 25.

    Mozaffarian, D. et al. Heart disease and stroke statistics-2016 update a report from the American Heart Association. Circulation 133, e38–e48 (2016).

    Google Scholar 

  26. 26.

    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).

    Article  Google Scholar 

  27. 27.

    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).

    Article  Google Scholar 

  28. 28.

    Zinkevich, M. Rules of machine learning: best practices for ML engineering (2017).

  29. 29.

    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).

    Article  Google Scholar 

  30. 30.

    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).

    Article  Google Scholar 

  31. 31.

    Lapuschkin, S. et al. Unmasking clever hans predictors and assessing what machines really learn. Nat. Commun. 10, 1096 (2019).

    Article  Google Scholar 

  32. 32.

    Pfungst, O. Clever Hans: (the Horse of Mr. Von Osten.) A Contribution to Experimental Animal and Human Psychology (Holt, Rinehart and Winston, 1911).

  33. 33.

    Machine Learning Recommendations for Policymakers (IIF, 2019); https://www.iif.com/Publications/ID/3574/Machine-Learning-Recommendations-for-Policymakers

  34. 34.

    Deeks, A. The judicial demand for explainable artificial intelligence. (2019).

  35. 35.

    Plumb, G., Molitor, D. & Talwalkar, A. S. Model agnostic supervised local explanations. Adv. Neural Inf. Process. Syst. 31, 2520–2529 (2018).

    Google Scholar 

  36. 36.

    Young, H. P. Monotonic solutions of cooperative games. Int. J. Game Theor. 14, 65–72 (1985).

    MathSciNet  Article  Google Scholar 

  37. 37.

    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).

  38. 38.

    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).

  39. 39.

    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).

  40. 40.

    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).

    Article  Google Scholar 

  41. 41.

    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).

    Article  Google Scholar 

Download references

Acknowledgements

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.

Author information

Affiliations

Authors

Contributions

S.M.L. and S.I.L conceived the study. S.M.L. designed algorithms, designed visualizations, designed metrics, ran experiments and contributed to the writing. G.E. ran experiments, designed visualizations and contributed to the writing. H.C. designed algorithms, ran experiments and contributed to the writing. A.D. performed dataset creation. R.K., J.H. and N.B. did dataset selection, model vetting and defined the chronic kidney disease prediction problem. J.M.P., B.N., R.K., J.H. and N.B. each contributed writing and helped procure and interpret datasets. S.-I.L. supervised research, method development and contributed to the writing.

Corresponding author

Correspondence to Su-In Lee.

Ethics declarations

Competing interests

The authors declare no competing interests.

Additional information

Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Supplementary information

Supplementary Information

Supplementary Figs, methods and references.

Supplementary Data 1

Rights and permissions

Reprints and Permissions

About this article

Verify currency and authenticity via CrossMark

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

Download citation

Further reading

Search

Quick links

Nature Briefing

Sign up for the Nature Briefing newsletter — what matters in science, free to your inbox daily.

Get the most important science stories of the day, free in your inbox. Sign up for Nature Briefing