.png){kind=icon}
返回
Star: semi-supervised tripartite attribute reduction
.png){kind=icon}
PRE
AI
DOI：10.1007/s13042-024-02472-1.png)
.png){kind=icon}
原文求助.png){kind=icon}
分享.png){kind=icon}
收藏
摘要
En 中文
Attribute reduction, also known as feature selection is favored in preprocessing data especially high-dimensional data for learning models. In general, conventional feature selection algorithms including supervised and unsupervised ones yield impressive performances depending on completely labeled and unlabeled data. Nevertheless, they fail for partially labeled data where a very limited portion is annotated with labels. In this work, we therefore propose a novel scheme dubbed as Semi-supervised Tripartite Attribute Reduction (STAR) to handle the paucity of label information. Essentially, STAR is a three-phase framework: (1) divide original partially labeled data into three parts including augmented labeled data, pseudo labeled data and updated unlabeled data; (2) devise three types of fuzzy measurements including fuzzy-rough dependency-based index, fuzzy joint entropy and fuzzy inter-cluster distance to eliminate feature quality on these three parts; (3) derive the qualified feature subset that can maximize the feature importance fusing the three evaluation criteria. STAR is validated in extensive experiments as compared with other nine well-established feature selection algorithms including one unsupervised, two supervised, and six semi-supervised methods. The reported results demonstrate that base classifiers fed by features selected from STAR are more accurate, suggesting its superiority in the presence of partially labeled data.
Keyword:
Attribute reduction
Fuzzy feature evaluation
Partially label data
Semi-supervised feature selection
期刊
IF:
2.7
论文数:
3.2K
被引数:
5.6K
机构
引用论文
Infrared Absorption Spectra of Alkali Metal Nitrates and Nitrites above and below the Melting Point
PREAI
分享收藏Semi-supervised feature selection by minimum neighborhood redundancy and maximum neighborhood relevancy.png)
APPLIED INTELLIGENCE
APPLIED INTELLIGENCEIF3.5
PREAI
分享收藏Fuzzy feature factorization machine: Bridging feature interaction, selection, and construction模糊特征分解机: 桥接特征交互，选择和构造
PREAI
分享收藏Maximal cliques-based hybrid high-dimensional feature selection with interaction screening for regressionNEUROCOMPUTING
NEUROCOMPUTINGIF6.5
PREAI
分享收藏
查看更多内容