A novel multi-objective forest optimization algorithm for wrapper feature selection

作者：

Highlights：

• A new multi-objective wrapper method based on Forest Optimization (MOFOA) is proposed.

• MOFOA uses archive, grid, and region-based selection to maintain Pareto front.

• Two continues and binary versions of MOFOA is presented to solve features selection.

• Continuous version of MOFOA outperforms other multi-objective algorithms.

• The performance of MOFOA was confirmed by quantitative and qualitative analyses.

摘要

•A new multi-objective wrapper method based on Forest Optimization (MOFOA) is proposed.•MOFOA uses archive, grid, and region-based selection to maintain Pareto front.•Two continues and binary versions of MOFOA is presented to solve features selection.•Continuous version of MOFOA outperforms other multi-objective algorithms.•The performance of MOFOA was confirmed by quantitative and qualitative analyses.

论文关键词：Feature selection,Multi-objective optimization,Forest optimization algorithm,Wrapper method,Dimension reduction

论文评审过程：Received 30 January 2020, Revised 6 December 2020, Accepted 14 February 2021, Available online 19 February 2021, Version of Record 11 March 2021.

论文官网地址：https://doi.org/10.1016/j.eswa.2021.114737