结合密度峰值和集成过滤器的自训练算法

doi:10.19504/j.cnki.issn1671-5365.2024.06.03

	结合密度峰值和集成过滤器的自训练算法
其他题名	Self-training Algorithm Combining Density Peak and Integrated Filter
	韩运龙; 尚庆生; 赵薇; 郭泓
	2023-12-06
发表期刊	宜宾学院学报
卷号	24 期号:6 页码:1-12
摘要	准确选取高置信度样本是提升自训练算法分类性能的关键。针对自训练迭代过程中的误分类样本，提出一种结合密度峰值和集成过滤器的自训练算法：利用密度峰值聚类计算样本的密度和峰值，构建初始高置信度样本集；为了过滤自训练迭代过程中的误分类样本，设计一个集成过滤器，从初始高置信度样本集进一步选择高置信度样本，将其添加进有标签样本集中迭代训练。在9个数据集上与4个相关的自训练算法进行对比实验，结果表明，算法的平均准确率和F分数分别为67.90%和65.54%，其分类性能显著优于对比算法。
其他摘要	Accurately selecting high confidence samples is the key to improve the classification performance of self-training algo-rithm.A self-training algorithm combining density peaks and integrated filters was proposed to address misclassified samples in self-training iteration process.The algorithm first used density peak clustering to calculate the density and peak value of samples,and constructed an initial high confidence sample set.Secondly,in order to filter out misclassified samples in self-training iteration process,a novel integrated filter was designed.High confidence samples were further selected from the initial high confidence sample set and added to the labeled sample set for iterative training.Comparative experiments were conducted with 4 related self-training algorithms on 9 datasets.The experimental results show that the average accuracy and F-score of the proposed algorithm are 67.90％and 65.54％respectively,and its classification performance is significantly superior to that of the comparison algorithm.
关键词	自训练无标签样本高置信度样本密度峰值集成过滤器
DOI	10.19504/j.cnki.issn1671-5365.2024.06.03
URL	查看原文
收录类别	AMI
ISSN	1671-5365
语种	中文
原始文献类型	学术期刊
中图分类号	TP18
CN号	51-1630/Z
文献类型	期刊论文
条目标识符	http://ir.lzufe.edu.cn/handle/39EH0E1M/35411
专题	信息工程与人工智能学院长青学院
通讯作者	尚庆生
作者单位	兰州财经大学信息工程学院
第一作者单位	信息工程与人工智能学院
通讯作者单位	信息工程与人工智能学院
推荐引用方式 GB/T 7714	韩运龙,尚庆生,赵薇,等. 结合密度峰值和集成过滤器的自训练算法[J]. 宜宾学院学报,2023,24(6):1-12.
APA	韩运龙,尚庆生,赵薇,&郭泓.(2023).结合密度峰值和集成过滤器的自训练算法.宜宾学院学报,24(6),1-12.
MLA	韩运龙,et al."结合密度峰值和集成过滤器的自训练算法".宜宾学院学报 24.6(2023):1-12.