All Models are Wrong, but Many are Useful: Learning a Variable's Importance by Studying an Entire Class of Prediction Models Simultaneously

DOI10.48550/arXiv.1801.01489zbMath1436.62019arXiv1801.01489MaRDI QIDQ97217

Francesca Dominici, Cynthia Rudin, Aaron Fisher

Publication date: 4 January 2018

Full work available at URL: https://arxiv.org/abs/1801.01489

zbMATH Keywords

U-statisticstransparencyinterpretable modelspermutation importanceconditional variable importanceRashomon

Mathematics Subject Classification ID

Inference from stochastic processes and prediction (62M20) Classification and discrimination; cluster analysis (statistical aspects) (62H30) Applications of statistics to social sciences (62P25) Foundations and philosophical topics in statistics (62A01) Order statistics; empirical distribution functions (62G30)

Related Items (24)

Visualizing Variable Importance and Variable Interaction Effects in Machine Learning Models ⋮ Scrutinizing XAI using linear ground-truth data with suppressor variables ⋮ Techniques to improve ecological interpretability of black-box machine learning models. Case study on biological health of streams in the United States with gradient boosted trees ⋮ Grouped feature importance and combined features effect plot ⋮ The explanation game: a formal framework for interpretable machine learning ⋮ Feature importance in neural networks as a means of interpretation for data-driven turbulence models ⋮ SLISEMAP: supervised dimensionality reduction through local explanations ⋮ Cross-model consensus of explanations and beyond for image classification models: an empirical study ⋮ Variable Selection Via Thompson Sampling ⋮ Visualizing the Implicit Model Selection Tradeoff ⋮ Machine learning meta-models for fast parameter identification of the lattice discrete particle model ⋮ Explaining classifiers with measures of statistical association ⋮ Improved feature selection with simulation optimization ⋮ Considerations when learning additive explanations for black-box models ⋮ Interpretable Architecture Neural Networks for Function Visualization ⋮ Prospects for Higgs boson and new scalar resonant production searches in \textit{ttbb} final state at the LHC ⋮ Explaining individual predictions when features are dependent: more accurate approximations to Shapley values ⋮ Testing conditional independence in supervised learning algorithms ⋮ Robust boosting for regression problems ⋮ iml ⋮ Unrestricted permutation forces extrapolation: variable importance requires at least one more model, or there is no free variable importance ⋮ Interpretable machine learning: fundamental principles and 10 grand challenges ⋮ Stochastic Tree Search for Estimating Optimal Dynamic Treatment Regimes ⋮ Structural importance and evolution: an application to financial transaction networks

Uses Software

• QCQP

Cites Work

• Unnamed Item
• Unnamed Item
• Unnamed Item
• Unnamed Item
• Unnamed Item
• Correlation and variable importance in random forests
• BART: Bayesian additive regression trees
• A new variable importance measure for random forests with missing data
• Matching methods for causal inference: a review and a look forward
• Unbiased split selection for classification trees based on the Gini index
• Empirical characterization of random forest variable importance measures
• Statistical modeling: The two cultures. (With comments and a rejoinder).
• Grouped variable importance with random forests and application to multiple functional data analysis
• DC programming: overview.
• On combining machine learning with decision making
• Fairness through awareness
• Comparing the Areas under Two or More Correlated Receiver Operating Characteristic Curves: A Nonparametric Approach
• Approximation Theorems of Mathematical Statistics
• Prediction uncertainty and optimal experimental design for learning dynamical systems
• Stability Selection
• Probability Inequalities for Sums of Bounded Random Variables
• On Nonlinear Fractional Programming
• A Survey of the S-Lemma
• A Class of Statistics with Asymptotically Normal Distribution
• Who Invented the Delta Method?
• Random forests

This page was built for publication: All Models are Wrong, but Many are Useful: Learning a Variable's Importance by Studying an Entire Class of Prediction Models Simultaneously