BMC Bioinformatics BioMed Central Methodology article Conditional variable importance for random forests

2008

Subfields: Evaluation & Benchmarks

Citations in the corpus

• Visualizing the Feature Importance for Black Box Models (2018, background): “ature importances according to the condition considered in the conditional PFI. The relevance of conditional feature importance in the case of random forests with correlated features was discussed in [20]. In Fig. 4, we illustrate the usefulness of a model-agnostic conditional PFI in case of interac”
• Explanations of model predictions with live and breakDown packages (2018, background): “In particular, decision trees which can be plotted using party package (Strobl et al., 2008) are well suited for this task, as they can help discover interactions.”
• Consistent Individualized Feature Attribution for Tree Ensembles (2018, methodology): “Different choices about the method of feature value permutation lead to variations of this basic approach [1, 10, 14, 23, 26].”
• Auditing black-box models for indirect influence (2016, background): “[8], a number of other works have looked at black-box auditing, primarily for direct influence [9], [27], [25], [24].”
• Interpreting random forest classification models using a feature contribution method (2013, background): “However, in [16], the authors showed that the above measures are biased in favor of continuous variables and variables with many categories.”
• Hyper-Parameter Optimization: A Review of Algorithms and Applications (2020, methodology): “TPE outperforms the Bayesian method in dealing with conditional variables because of its tree structure (Strobl et al., 2008).”
• All Models are Wrong, but Many are Useful: Learning a Variable’s Importance by Studying an Entire Class of Prediction Models Simultaneously (2018, methodology): “This RF VI measure has been the topic of empirical studies (Archer and Kimes, 2008; Calle and Urrea, 2010; Wang et al., 2016), and several variations of the measure have been proposed (Strobl et al., 2007, 2008; Altmann et al., 2010; Hapfelmeier et al., 2014).”
• How does forest structure affect root reinforcement and susceptibility to shallow landslides? (2016, methodology): “For the multivariate analysis, we used logistic regression (Hosmer and Lemeshow, 2000) and random forest (Breiman, 2001; Strobl et al., 2008; Strobl et al., 2009a).”
• Fifty Years of Classification and Regression Trees (2014, background): “Strobl et al. (2007b) showed that the scores are biased toward correlated variables, and Strobl et al. (2008) proposed an alternative permutation scheme as a solution.”
• Integrating ensemble species distribution modelling and statistical phylogeography to inform projections of climate change impacts on species distributions (2013, methodology): “integrifolia, while controlling for the highly correlated nature of these predictors (Strobl et al. 2008).”
• Understanding Global Feature Contributions With Additive Importance Measures (2020, methodology): “Since permutation tests break feature dependencies, one variation suggests using a conditional permutation scheme [36].”
• Revisiting acoustic correlates of pharyngealization in Jordanian and Moroccan Arabic: Implications for formal representations (2017, methodology): “Strobl et al. (2008, 2007) found that the original randomForest provided biased estimates of Random Forests’ variable importance (see below) as it was biased towards variables with multiple categories and multiple cut-points, and also overestimated variable importance measures when correlated data…”
• Explaining by Removing: A Unified Framework for Model Explanation (2020, methodology): “…to how Occlusion, PredDiff and CXPlain calculate local feature importance and how permutation tests and feature ablation (LOCO) summarize global feature importance (Breiman, 2001; Strobl et al., 2008; Zeiler and Fergus, 2014; Zintgraf et al., 2017; Lei et al., 2018; Schwab and Karlen, 2019).”