一种提高SVM分类速度和泛化性的新方法

A New Method to Advance Efficiency and Generalization of SVM

提出了一种改进的支持向量分类方法,根据支持向量机中支持向量不会出现在两类样本集间隔以外的正确划分区的理论,通过引入类质心,类半径,类质心距等概念,从而较好地解决快速而准确的删除非支持向量的问题,引入了类向心度的概念,解决了当两类样本集混淆严重的时候如何更加精确的进行剔除混淆点,保证算法泛化性的问题。实验表明,采用这种改进的算法既能快速精确的对训练样本进行删减又可以当两类训练样本集混淆较严重时较好的解决泛化性问题。

An improved SVM algorithm is proposed based on the theory that support vector will not appears in the areas which out of the interval between two classes. Benefit from the concepts of class-radius, class-centroid-distance and class-centripetal force et al. , we can delete those non-SV effectively with high accuracy and generality even the data was promiscuous. The experiments show that, comparing with other algorithms, our method achieved a satisfactory result.