一种改进的支持向量机降维方法

IMPROVED DIMENSIONALITY REDUCTION METHOD BASED ON SUPPORT VECTOR MACHINES

提出了一种基于支持向量机的改进的降维方法。在输入和特征空间中,特征子集的选取分别根据原始特征每一维对分类的贡献来获得。最后,通过将输入和特征空间中的特征选取联合起来,得到了一种改进的降维方法。实验表明:使用这种方法,在保持对分类准确率不受明显的影响的同时,能大大地提高训练和预测的速度。

In this paper we present an improved dimensionality reduction method based on support vector machines （SVMs）. In both input and feature space, a subset of features was selected by ranking its contributions to the classification associated to its original features respectively. Accordingly, we developed an improved dimensionality reduction method by the combination of the feature selections in input and feature space. Experiments showed that training SVMs to use the selected subset features, which were obtained by our method, was helpful to improve the training and prediction speed without a significant loss in classification performance.