Oblique predictive clustering trees
作者：
Highlights：
• Ensembles of predictive clustering trees scale poorly with the number of targets.
• Oblique predictive clustering trees use linear combinations of features in the splits.
• The proposed methods retain or improve the premium accuracy of ensembles of trees.
• Our methods scale much better to high dimensional outputs and exploit sparse data.
• Meaningful feature importance scores can be extracted from ensembles of oblique trees.
摘要
•Ensembles of predictive clustering trees scale poorly with the number of targets.•Oblique predictive clustering trees use linear combinations of features in the splits.•The proposed methods retain or improve the premium accuracy of ensembles of trees.•Our methods scale much better to high dimensional outputs and exploit sparse data.•Meaningful feature importance scores can be extracted from ensembles of oblique trees.
论文关键词：Oblique decision trees,Predictive clustering trees,Ensembles,Structured output prediction,Sparse data
论文评审过程：Received 28 July 2020, Revised 14 April 2021, Accepted 9 June 2021, Available online 12 June 2021, Version of Record 22 June 2021.
论文官网地址：https://doi.org/10.1016/j.knosys.2021.107228
