.png){kind=icon}
返回
Recursive elimination-election algorithms for wrapper feature selection
.png){kind=icon}
PRE
AI
DOI：10.1016/j.asoc.2021.107956.png)
.png){kind=icon}
原文求助.png){kind=icon}
分享.png){kind=icon}
收藏
摘要
En 中文
For classification tasks in machine learning, this paper proposes a brand-new wrapper feature selection algorithm prototype named recursive elimination-election (REE), which is conceived in a simple but exquisite structure inspired by the recursion technique in computer science. Prevalent metaheuristic methods such as differential evolution (DE), particle swarm optimization (PSO), etc., from evolutionary computation (EC) and swarm intelligence (SI) communities have recently been widely applied to feature selection research, but suffer from severe drawbacks including but not limited to low efficient binary representation transformation, poor population diversity, excessive control parameter adjustments and sophisticated mechanisms. Instead, REE is organically constructed with an ordinary subset representation of feature indexes, simple operators, getting rid of extra control parameters. Specifically, REE is assembled of two basic recursive sub-algorithms, i.e., recursive random bisection elimination (RRBE) and recursive greedy binary election (RGBE), which somewhat embody the idea of divide-and-conquer. By inspecting smaller and potential feature subsets in recursive ways, better subsets are returned automatically. A comprehensive experimental study was conducted on 14 UCI and ASU benchmark datasets with feature sizes ranging from dozens to thousands by using REE together with 6 state-of-the-art metaheuristic algorithms for comparison. The results show that the proposed REE has competitive search ability for feature selection problems, and it is especially prominent in handling high-dimensional datasets. Therefore, REE is promising to become a wrapper feature selection search paradigm with low solution cost and high efficiency. (C) 2021 Elsevier B.V. All rights reserved.
Keyword:
Wrapper feature selection
Classification
Recursion technique
High dimensionality
期刊
IF:
6.6
论文数:
1.4W
被引数:
4.8W
机构
暂无机构信息
引用论文
分享收藏Binary dragonfly optimization for feature selection using time-varying transfer functions使用时变传递函数进行特征选择的二进制蜻蜓优化
OAAI
OAAI分享收藏Variable-Length Particle Swarm Optimization for Feature Selection on High-Dimensional Classification高维分类特征选择的变长粒子群算法OAAI
OAAI分享收藏A genetic algorithm-based method for feature subset selection一种基于遗传算法的特征子集选择方法.png)
SOFT COMPUTING
SOFT COMPUTINGIF2.5
PREAI
分享收藏分享收藏Mechanism of insulin incorporation into .alpha.2-macroglobulin: implications for the study of peptide and growth factor bindingBiochemistry
BiochemistryIF0
PREAI
分享收藏Chaotic dragonfly algorithm: an improved metaheuristic algorithm for feature selection混沌蜻蜓算法: 一种改进的元启发式特征选择算法APPLIED INTELLIGENCEOAAI
APPLIED INTELLIGENCEIF3.5
OAAI分享收藏
查看更多内容