郑州大学学报（理学版）2016，Vol.48Issue(3)：57-62,6.DOI:10.13705/j.issn/1671-6841.2016087
基于局部邻域嵌入的无监督特征选择
Unsupervised Feature Selection Based on Local Neighborhood Embedding
摘要
Abstract
In machine learning, feature selection can effectively reduce the data dimension. Manifold learning and l2,1 norm can improve the effectiveness and efficiency of feature selection. Manifold learning can maintain the geometric structure of the original data. l2,1 norm can prevent the over fitting, and it can enhance the generalization ability of the model. A new unsupervised feature selection method based on lo-cal neighborhood embedding ( LNE) algorithm and l2,1 norm were proposed. The main ideas were as fol-lowing. Firstly, it constructed similar matrix by the distances and reconstruction coefficients between each data and its neighborhood. Secondly, it built a low dimensional space and used l2,1 norm sparse regres-sion. Finally, it calculated the importance of each feature and chose the optimal feature subset. Experi-mental results showed the effectiveness of the proposed algorithm by comparing it with several typical fea-ture selection algorithms.关键词
机器学习/局部邻域嵌入/流形学习/无监督特征选择Key words
machine learning/local neighborhood embedding/manifold learning/unsupervised feature selection分类
信息技术与安全科学引用本文复制引用
脱倩娟,赵红..基于局部邻域嵌入的无监督特征选择[J].郑州大学学报（理学版）,2016,48(3):57-62,6.基金项目
国家自然科学基金资助项目(61379049,61472406) （61379049,61472406）
福建省自然科学基金资助项目(2015J01269) （2015J01269）
漳州市自然科学基金资助项目(ZZ2016J35) （ZZ2016J35）
