《计算机应用研究》|Application Research of Computers

防止过拟合的属性约简

Attribute reduction with avoiding overfitting

免费全文下载 (已被下载 次)  
获取PDF全文
作者 沈微微,李颖,杨志豪,王祥力,叶轩
机构 宿迁学院 信息工程学院,江苏 宿迁 223800
统计 摘要被查看 次,已被下载
文章编号 1001-3695(2020)09-020-2665-04
DOI 10.19734/j.issn.1001-3695.2019.05.0116
摘要 近年来,机器学习的过拟合问题备受关注,尤其在属性约简中。为解决这一难题,提出一种融合集成策略和去除操作的算法。首先将训练模型数据平分为<i>M份;然后将其中M</i>-1份采用集成策略进行潜在约简计算;最后将剩余的一份进行提前测试,一旦发生过拟合则将刚加入的属性从潜在约简集中去除。利用提前测试潜在属性约简的方法来防止过拟合现象的发生,几组UCI数据的实验结果说明了新算法的有效性,同时为丰富和发展属性约简提供了一种新的方向。
关键词 粗糙集; 过拟合; 集成策略; 属性约简
基金项目 国家自然科学基金资助项目(61871228)
国家青年基金资助项目(61702322)
江苏省高校自然科学基金资助项目(18KJB520049)
宿迁市科技计划资助项目(Z2019109)
本文URL http://www.arocmag.com/article/01-2020-09-020.html
英文标题 Attribute reduction with avoiding overfitting
作者英文名 Shen Weiwei, Li Ying, Yang Zhihao, Wang Xiangli, Ye Xuan
机构英文名 School of Information Engineering,Suqian College,Suqian Jiangsu 223800,China
英文摘要 Nowadays, overfitting is becoming a more and more important issue in the aspect of machine learning, especially in attribute reduction, the problem of overfitting has caused a lot of trouble. In order to solve this problem, this paper proposed an algorithm that integrated ensemble strategy and removal operation. First of all, this algorithm divided the data sets into <i>M</i> groups in the same size. Then, it adopted the ensemble strategy to potential attribute reduction for the <i>M</i>-1 groups. Finally, it tested the remaining 1 group in advance. Once overfitting had occurred, it removed the newly joined properties from the potential attribute reduction. It used this method to test potential attributes in advance to prevent overfitting in the actual application. The experimental results of several UCI datasets show the validity of the new algorithm, which may provide a new direction for enriching and developing attribute reduction.
英文关键词 rough set; overfitting; ensemble strategy; attribute reduction
参考文献 查看稿件参考文献
 
收稿日期 2019/5/14
修回日期 2019/7/10
页码 2665-2668
中图分类号 TP391
文献标志码 A