Dependence-biased clustering for variable selection with random forests

作者：

Highlights：

• We introduce a novel conditional permutation measure for variable importance.

• This measure leverages inter-variable dependencies via biased K-means clustering.

• This measure allows to select a small number of relevant, non-redundant variables.

• Extensive results show our variable selection approach is very effective in practice.

摘要

•We introduce a novel conditional permutation measure for variable importance.•This measure leverages inter-variable dependencies via biased K-means clustering.•This measure allows to select a small number of relevant, non-redundant variables.•Extensive results show our variable selection approach is very effective in practice.

论文关键词：Variable selection,Random forest,Permutation importance,Regression,Classification,Clustering

论文评审过程：Received 13 October 2018, Revised 29 May 2019, Accepted 21 July 2019, Available online 24 July 2019, Version of Record 5 August 2019.

论文官网地址：https://doi.org/10.1016/j.patcog.2019.106980