基于特征空间中类别可分性判据的特征选择

Feature Selection based on Separability in Feature Space

针对支持向量机首先要将数据通过一个非线性函数映射到高维特征空间,从而在特征空间中改变了数据的分布,因此在设计支持向量机分类器时有必要在特征空间进行特征选择,而非原空间。提出在特征空间中利用类别可分性判据进行特征选择,将类内类间距离引申到特征空间来计算可分性判据。通过仿真,该方法能够有效地在特征空间进行特征选择。

The data were mapped into a higher dimensional feature space by a nonlinear function in the support vector machine.So the distribution of the data was changed in feature space,and the feature selection must be done in the feature space,but not in the primal space.The feature selection that based on separability in feature space was proposed in this paper.The distances that inter-class and inner-class were introduced into the feature space,so that the separability criterion can be calculated in feature space.And the Separability criterion was used to select feature in feature space.The method was confirmed effective by simulation.