基于多区域划分的模糊支持向量机方法

摘要

Considering that current fuzzy support vector machine(FSVM) can’t effectively locate support vectors and thus causes loss of classified information, a FSVM method based on multi-region partition was proposed. As approximate estimation of FSVM support vectors, the general position of support vectors was obtained with traditional SVM, which further fuses the support vector domain description with negative examples(SVDD-neg) model to divide the whole sample space. The results show that the fuzzy membership is determined according to the position of samples in space by different rules, which weakens the outliers and increases the membership of support vectors. The proposed method is more robust and can gain a better generalization ability compared with the FSVM based on affinity among samples and the FSVM based on the distance between a sample and its hyperplane within the class.%针对模糊支持向量机(FSVM)方法无法有效定位支持向量，在确定隶属度时易丢失分类信息的问题，提出一种基于多区域划分的FSVM方法。该方法先利用传统SVM获取支持向量的大体位置，作为对FSVM支持向量的近似估计，再进一步融合带负类样本的支持向量域描述(SVDD-neg)模型，对样本空间进行划分，最后根据样本所在的区域按不同的规律确定隶属度。研究结果表明：这种隶属度确定方式不仅能有效削弱野值样本的影响，而且也会提高支持向量的隶属度。与基于样本紧密度以及基于样本到类内超平面距离的FSVM方法相比，该方法具有更好的抗噪性能和泛化能力。