北京邮电大学学报

  • EI核心期刊

北京邮电大学学报 ›› 2021, Vol. 44 ›› Issue (4): 49-55.doi: 10.13190/j.jbupt.2020-237

• 论文 • 上一篇    下一篇

粒子群优化的模糊粗糙集双约简算法

刘占峰, 潘甦   

  1. 南京邮电大学 江苏省通信与网络技术工程研究中心, 南京 210003
  • 收稿日期:2020-11-23 发布日期:2021-07-13
  • 作者简介:刘占峰(1980-),男,博士生,E-mail:zf.liu@139.com;潘甦(1969-),男,教授,博士生导师.
  • 基金资助:
    江苏省研究生科研与实践创新计划项目(KYCX18_0882);国家自然科学基金项目(6201244);江苏省重点研发计划项目(BE2018733)

Fuzzy-Rough Bireducts Algorithm Based on Particle Swarm Optimization

LIU Zhan-feng, PAN Su   

  1. Jiangsu Engineering Research Center of Communication and Network Technology, Nanjing University of Posts and Telecommunications, Nanjing 210003, China
  • Received:2020-11-23 Published:2021-07-13

摘要: 为了提升下游模型的性能,获得质量更好的约简数据集,提出基于粒子群优化(PSO)的模糊粗糙集特征和实例联合选择算法,引入基于ε-双约简的适应度函数来评估约简集的质量,引导搜索过程快速逼近最优解.实验结果表明,基于PSO算法的模糊粗糙集双约简算法有效约简了实例和特征,获得了高质量的约简集,在分类任务中取得了优于原始数据集的准确度.

关键词: 模糊粗糙集, 特征选择, 实例选择, 粒子群优化

Abstract: Selecting informative features and removing noise instances are beneficial to gain a clean dataset and promote the performance of subsequent classifiers. A novel algorithm for fuzzy-rough bireducts with particle swarm optimization is proposed. The fitness function with ε-bireduct is employed to evaluate the candidate fuzzy-rough bireducts, which drives the particle swarm optimization search process toward better candidate solutions. The selected optimal bireduct is utilized to construct the subsequent classifier. The experimental results show that the proposed algorithm is superior to the counterpart, which reduces the instances and features effectively, and obtains high-quality bireducts. The classification accuracy of the proposed algorithm is thus better than the counterpart.

Key words: fuzzy-rough, feature selection, instance selection, particle swarm optimization

中图分类号: