High-dimensional hybrid feature selection using interaction information-guided search

作者：

Highlights：

• A new high-dimensional hybrid feature selection algorithm called IGIS is proposed.

• Interaction information is employed to guide the search.

• Our method is dynamic and selects only relevant and irredundant features.

• The experimental results show that IGIS outperforms prior wrapper methods.

摘要

•A new high-dimensional hybrid feature selection algorithm called IGIS is proposed.•Interaction information is employed to guide the search.•Our method is dynamic and selects only relevant and irredundant features.•The experimental results show that IGIS outperforms prior wrapper methods.

论文关键词：Feature selection,High-dimensional data sets,Hybrid algorithms,Interaction information,Sequential search

论文评审过程：Received 11 May 2017, Revised 28 December 2017, Accepted 1 January 2018, Available online 2 January 2018, Version of Record 20 February 2018.

论文官网地址：https://doi.org/10.1016/j.knosys.2018.01.002