摘要
提出一种新颖的基于空间覆盖的半监督特征选择方法。该算法同时利用已标签数据与未标签数据进行特征选择,各特征的相关性大小由其在不同簇中的覆盖程度衡量。在公共数据集和毒性数据集上的实验表明,该方法在改善学习精度上有很好的应用前景。
A novel Spatial Overlapping based Semi-supervised Feature Selection (SOS-FS) method is proposed.It uses both labeled and unlabeled data in feature selection,feature's relevance is measured by its overlapping ratio among different clusters. Experimental results carried out on some public datasets collected from the UCI machine learning repository and predictive toxicology domain show that SOS-FS has a promising performance on the improvement of the learning accuracy.
出处
《计算机工程与应用》
CSCD
北大核心
2010年第8期130-132,140,共4页
Computer Engineering and Applications
基金
福建省自然科学基金 Grant No.2007J0016~~
关键词
空间覆盖
半监督
特征选择
spatial overlapping
semi-supervised
feature selection