基于模糊隶属度的支持向量机去噪方法

针对传统支持向量机对于噪声和野点敏感的问题,采用一种模糊技术去除样本中的噪声和野点.应用基于样本之间的紧密度确定每个样本的模糊隶属度,通过训练确定阀值,去除影响得到最优分类超平面的噪声和野点.实验结果表明,与传统的支持向量机相比,该方法提高了支持向量机的抗噪能力,在不影响精度的前提下,线性规划下的一类分类方法要比二次规划节省很多时间.

Approach of removing noises and outliers for SVM based on fuzzy membership

A fuzzy technology is used to remove noises and outliers,for the question that traditional support vector machine is sensitive to noises and outliers.The fuzzy membership of each sample is defined by affinity among samples,and by the training determine a threshold,noises and outliers are removed,which influence optimal separating hyperplane.Experimental results show that this approach has made a better effect on improving the anti-noise capability of SVM.On the premise of without affecting the accuracy,the ...