基于核空间类间平均距的径向基函数—支持向量机特征选择算法

SVM-RFE特征选择算法的算法复杂度高,特征选择消耗时间过长,为了缩短特征选择的时间,针对径向基函数—支持向量机分类器提出了依据核空间类间平均距进行特征选择的算法。首先分析了径向基函数核参数与数据集核空间类间平均距之间的关系,然后提出了依据单个特征对数据集的核空间类间平均距的贡献大小进行特征重要性排序的算法,最后用该算法和SVM-RFE算法分别对8个UCI数据集进行了特征选择实验。实验结果证明了该算法的正确性、有效性,而且特征选择的时间与SVM-RFE算法相比大大减小。

RBF-SVM feature selection arithmeticbased on kernel space mean inter-class distance

The feature selection arithmetic of SVM-RFE is complex, and it always costs much time to select feature. This paper put forward an arithmetic based on RBF-SVM which select feature through kernel space mean inter-class distance to save feature selecting time. Firstly, this paper analysed the relation between the kernel parameter and kernel space mean inter-class distance. Then, it put forward the arithmetic which order the features through every feature's contribution to the kernel space mean inter-class distance. Finally, it made feaure selection experiments with eight subsets of UCI. The results of experiments indicates that the arithmetic of this paper is right and usefull, and it costs less time than SVM-RFE.