An Efficient Explanation of Individual Classifications using Game Theory

By Erik Strumbelj, Igor Kononenko. 2010

802 citations recorded by the source corpus.

Subfields: Artificial Intelligence, Safety, Robustness & Interpretability, Game & Decision Theory

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

In Safety, Robustness & Interpretability

In Game & Decision Theory

Citations in the corpus

Anchors: High-Precision Model-Agnostic Explanations (2018, methodology): “…machine learning models (Lakkaraju, Bach, and Leskovec 2016; Ustun and Rudin 2015; Wang and Rudin 2015) to local explanations (for individual predictions) that can be computed for any classifier (Baehrens et al. 2010; Ribeiro, Singh, and Guestrin 2016b; Strumbelj and Kononenko 2010).”
Visualizing the Feature Importance for Black Box Models (2018, methodology): “A method that measures feature effects based on the Shapley value [19] from coalitional game theory was first presented for classification in [21] and has been extended to regression and global analysis in [22].”
A Survey of Methods for Explaining Black Box Models (2018, methodology): “Then, in Reference [29], Strumbelj et al.”
Model-Agnostic Interpretability of Machine Learning (2016, methodology): “In this section, we make a case for model-agnostic interpretability, as opposed to just using interpretable models.”
Interpreting individual classifications of hierarchical networks (2013, background): “In an attempt to provide such explanations, recent research has treated the entire classification system as a single black box in order to provide classifier-independent explanations [7].”
DALEX: explainers for complex predictive models (2018, methodology): “The most known approaches to explanations of a single prediction are LIME method (Ribeiro et al., 2016), working best for local explanations, and Shapley values (Štrumbelj and Kononenko, 2010, 2014; Lundberg and Lee, 2017), working best for variable attribution.”
Practical guide to SHAP analysis: Explaining supervised machine learning model predictions in drug development (2024, positive): “25 The authors showed how Shapley values can be used to fairly quantify the contribution of features in a ML model.”
Explaining Deep Neural Networks and Beyond: A Review of Methods and Applications (2021, methodology): “When transposing the method to the task of explaining an ML model [116], [179], the players of the cooperating game become the input features, and the payoff function becomes related to the DNN output.”
Understanding Global Feature Contributions With Additive Importance Measures (2020, methodology): “To efﬁciently generate subsets of features from the appropriate distribution, we enumerate over random permutations of the indices D = { 1 , . . . , d } , similar to Štrumbelj and Kononenko [18].”
Improving the Interpretability of Deep Neural Networks with Knowledge Distillation (2018, background): “Past research focus on either global interpretations [22] [23] or local explanations [24] [25] [26].”
Efficiently Explaining Decisions of Probabilistic RBF Classification Networks (2011, methodology): “Fortunately, this problem can be viewed from the point of coalitional game theory [12].”
Explaining machine learning models in sales predictions (2017, methodology): “Recently, two general methods for explaining classiﬁcation models and their predictions have been introduced (Robnik-ˇSikonja and Kononenko, 2008; ˇStrumbelj and Kononenko, 2010).”
Explanation and Reliability of Individual Predictions (2013, methodology): “However, several widely-used machine learning models are not additive, which leads to misleading and incorrect explanations of the importance and influence of features [29].”
A Survey on Neural Network Interpretability (2020, background): “Štrumbelj and Kononenko [72] approximate Shapley value by sampling from S(N)×X (Cartesian product).”
SurvNAM: The machine learning survival model explanation (2021, methodology): “Another powerful method is the SHAP [30] which takes a game-theoretic approach for optimizing a regression loss function based on Shapley values [31].”