%0 Journal Article %1 strobl_conditional_2008 %A Strobl, Carolin %A Boulesteix, Anne-Laure %A Kneib, Thomas %A Augustin, Thomas %A Zeileis, Achim %D 2008 %J BMC Bioinformatics %K Decision Forest, Random Trees, conditional correlations, importance, multicollinearity, relationships, selection variable %N 1 %P 307 %R 10.1186/1471-2105-9-307 %T Conditional variable importance for random forests %U http://www.biomedcentral.com.proxy.lib.uiowa.edu/1471-2105/9/307/abstract %V 9 %X Background Random forests are becoming increasingly popular in many scientific fields because they can cope with "small n large p" problems, complex interactions and even highly correlated predictor variables. Their variable importance measures have recently been suggested as screening tools for, e.g., gene expression studies. However, these variable importance measures show a bias towards correlated predictor variables. Results We identify two mechanisms responsible for this finding: (i) A preference for the selection of correlated predictors in the tree building process and (ii) an additional advantage for correlated predictor variables induced by the unconditional permutation scheme that is employed in the computation of the variable importance measure. Based on these considerations we develop a new, conditional permutation scheme for the computation of the variable importance measure. Conclusion The resulting conditional variable importance reflects the true impact of each predictor variable more reliably than the original marginal approach.

@article{strobl_conditional_2008, abstract = {Background Random forests are becoming increasingly popular in many scientific fields because they can cope with "small n large p" problems, complex interactions and even highly correlated predictor variables. Their variable importance measures have recently been suggested as screening tools for, e.g., gene expression studies. However, these variable importance measures show a bias towards correlated predictor variables. Results We identify two mechanisms responsible for this finding: (i) A preference for the selection of correlated predictors in the tree building process and (ii) an additional advantage for correlated predictor variables induced by the unconditional permutation scheme that is employed in the computation of the variable importance measure. Based on these considerations we develop a new, conditional permutation scheme for the computation of the variable importance measure. Conclusion The resulting conditional variable importance reflects the true impact of each predictor variable more reliably than the original marginal approach.}, added-at = {2017-01-09T13:57:26.000+0100}, author = {Strobl, Carolin and Boulesteix, Anne-Laure and Kneib, Thomas and Augustin, Thomas and Zeileis, Achim}, biburl = {https://www.bibsonomy.org/bibtex/24f0dc72d47c5e502bf347611444d7e99/yourwelcome}, doi = {10.1186/1471-2105-9-307}, interhash = {b9365c9d90fd6d6ffbd757665c89b68e}, intrahash = {4f0dc72d47c5e502bf347611444d7e99}, issn = {1471-2105}, journal = {BMC Bioinformatics}, keywords = {Decision Forest, Random Trees, conditional correlations, importance, multicollinearity, relationships, selection variable}, number = 1, pages = 307, timestamp = {2017-01-09T14:01:11.000+0100}, title = {Conditional variable importance for random forests}, url = {http://www.biomedcentral.com.proxy.lib.uiowa.edu/1471-2105/9/307/abstract}, urldate = {2013-06-28}, volume = 9, year = 2008 }