Heterogeneous oblique random forest

作者：

Highlights：

• We propose a heterogeneous oblique random forest that employ a linear (oblique) hyperplane at each node.

• The hyperplanes are obtained via linear classifiers trained on selected one-vs-all and hyperclasses based partitions.

• The selection of an oblique hyperplane at each node is based on the optimization of an impurity criterion.

• The heterogeneous oblique decision trees are more accurate and diverse than other standard decision tree variants.

• On benchmarking 190 classifiers on 121 UCI datasets, the proposed oblique random forests are the top 3 ranked classifiers.

摘要

•We propose a heterogeneous oblique random forest that employ a linear (oblique) hyperplane at each node.•The hyperplanes are obtained via linear classifiers trained on selected one-vs-all and hyperclasses based partitions.•The selection of an oblique hyperplane at each node is based on the optimization of an impurity criterion.•The heterogeneous oblique decision trees are more accurate and diverse than other standard decision tree variants.•On benchmarking 190 classifiers on 121 UCI datasets, the proposed oblique random forests are the top 3 ranked classifiers.

论文关键词：Benchmarking,Classifiers,Oblique random forest,Heterogeneous,One-vs-all,Ensemble learning

论文评审过程：Received 27 November 2018, Revised 2 August 2019, Accepted 12 October 2019, Available online 13 October 2019, Version of Record 21 October 2019.

论文官网地址：https://doi.org/10.1016/j.patcog.2019.107078