摘要
针对基于约束得分的特征选择容易受成对约束的组成和基数影响的问题,提出了一种基于约束得分的动态集成选择算法(dynamic ensemble selection based on bagging constraint score,BCS-DES)。该算法将bagging约束得分(bagging constraint score,BCS)引入动态集成选择算法,通过将样本空间划分为不同的区域,使用多种群并行遗传算法为不同测试样本选择局部最优的分类集成,达到提高分类精度的目的。在UCI实验数据集上进行的实验表明,BCS-DES算法较现有的特征选择算法受成对约束组成和基数影响更小,效果更好。
Aiming at the problem that the feature selection based on constraint score can be easily affected by the composition and eardinality of pairwise constraints, this paper presented a new method called dynamic ensemble selection based on bagging constraint score. The algorithm introduced bogging constraint score into dynamic ensemble selection, divided the sample space into different parts, and then used the multi：population genetic algorithm to select the optimal multi-classifiers ensemble for the accuracy of the local classification. The experimental results on UCI datasets illustrates that the BCS-DES is smaller affected by the composition and cardinality of the pairwise constraints than the current feature selection methods, and can get better results.
出处
《计算机应用研究》
CSCD
北大核心
2014年第3期708-712,共5页
Application Research of Computers
基金
国家自然科学基金资助项目(61004069)
安徽省自然科学基金资助项目(1208085QF107)
关键词
约束得分
动态集成选择
特征选择
分类器集成
成对约束
constraint score
dynamic ensemble selection
feature selection
classifier ensemble
pairwise constraints
