Consistent feature attribution for tree ensembles

By Scott M. Lundberg, Su-In Lee. 2017

176 citations recorded by the source corpus.

Note that a newer expanded version of this paper is now available at: arXiv:1802.03888 It is critical in many applications to understand what features are important for a model, and why individual predictions were made. For tree ensemble methods these questions are usually answered by attributing importance values to input features, either globally or for a single prediction. Here we show that current feature attribution methods are inconsistent, which means changing the model to rely more on a given feature can actually decrease the importance assigned to that feature. To address this problem we develop fast exact solutions for SHAP (SHapley Additive exPlanation) values, which were recently shown to be the unique additive feature attribution method based on conditional expectations that is both consistent and locally accurate. We integrate these improvements into the latest version of XGBoost, demonstrate the inconsistencies of current methods, and show how using SHAP values results in significantly improved supervised clustering performance. Feature importance values are a key part of understanding widely used models such as gradient boosting trees and random forests, so improvements to them have broad practical implications.

Subfields: Artificial Intelligence, Nature and Landscape Conservation, Statistics and Probability

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

1. Maarten de Rijke
2. Pavel Serdyukov
3. Yury Ustinovsky
4. Boris Sharchilev
5. W. Bruce Croft
6. Hamed Zamani

In Nature and Landscape Conservation

1. Maarten de Rijke
2. Pavel Serdyukov
3. Yury Ustinovsky
4. Boris Sharchilev

In Statistics and Probability

1. Maarten de Rijke
2. Pavel Serdyukov
3. Yury Ustinovsky
4. Boris Sharchilev

Citations in the corpus

• Finding Influential Training Samples for Gradient Boosted Decision Trees (2018, methodology): “…setups (Palczewska et al., 2013; Tolomei et al., 2017; Fong & Vedaldi, 2017) and interpretation methods, both model-agnostic (Ribeiro et al., 2016; Lundberg & Lee, 2017) and model-specific (Shrikumar et al., 2017; Tolomei et al., 2017; Sundararajan et al., 2017), have recently been proposed in the…”
• Explaining individual predictions when features are dependent: More accurate approximations to Shapley values (2019, unknown): “The skill score for method q takes the form Skill ( MAE, method q ) MAE ( method q ) MAE ( original ) MAE ( optimal ) MAE ( original ) ,”