Tree-based machine learning models such as random forests, decision trees, and gradient boosted trees are popular non-linear 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) 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. 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 i) identify high magnitude but low frequency non-linear mortality risk factors in the US population, ii) highlight distinct population subgroups 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.

Although anaesthesiologists strive to avoid hypoxemia during surgery, reliably predicting future intraoperative hypoxemia is not currently possible. Here, we report the development and testing of a machine-learning-based system that, in real time during general anaesthesia, predicts the risk of hypoxemia and provides explanations of the risk factors. The system, which was trained on minute-by-minute data from the electronic medical records of over fifty thousand surgeries, improved the performance of anaesthesiologists when providing interpretable hypoxemia risks and contributing factors. The explanations for the predictions are broadly consistent with the literature and with prior knowledge from anaesthesiologists. Our results suggest that if anaesthesiologists currently anticipate 15% of hypoxemia events, with this system’s assistance they would anticipate 30% of them, a large portion of which may benefit from early intervention because they are associated with modifiable factors. The system can help improve the clinical understanding of hypoxemia risk during anaesthesia care by providing general insights into the exact changes in risk induced by certain patient or procedure characteristics.

Interpreting predictions from tree ensemble methods such as gradient boosting machines and random forests is important, yet feature attribution for trees is often heuristic and not individualized for each prediction. Here we show that popular feature attribution methods are inconsistent, meaning they can lower a feature's assigned importance when the true impact of that feature actually increases. This is a fundamental problem that casts doubt on any comparison between features. To address it we turn to recent applications of game theory and develop fast exact tree solutions for SHAP (SHapley Additive exPlanation) values, which are the unique consistent and locally accurate attribution values. We then extend SHAP values to interaction effects and define SHAP interaction values. We propose a rich visualization of individualized feature attributions that improves over classic attribution summaries and partial dependence plots, and a unique "supervised" clustering (clustering based on feature attributions). We demonstrate better agreement with human intuition through a user study, exponential improvements in run time, improved clustering performance, and better identification of influential features. An implementation of our algorithm has also been merged into XGBoost and LightGBM, see http://github.com/slundberg/shap for details.