自适应模糊支持向量机算法

An Adaptive Fuzzy Support Vector Machine Algorithm

支持向量机算法对噪声和异常点是敏感的,为了克服这个问题,人们引入了模糊隶属度。传统确定样本模糊隶属度的方法,都是基于原始空间的。文章提出了基于特征空间的模糊隶属度函数模型。在该模型中,以特征空间中的样本为中心,以给定的距离d为半径作超球,根据其它样本落到超球内的个数来确定中心样本点的模糊隶属度。并将新的模糊隶属度模型引入自适应支持向量机,提出了模糊自适应支持向量机算法。实验结果表明,该模型能有效地提高自适应支持向量机的抗噪能力和预测精度。

Support Vector Machine （SVM） is sensitive to the noises and outliers.To overcome this drawback,fuzzy membership function is introduced,which is based on the primal space in the traditional method.In this paper,a novel fuzzy membership function model is presented based on the feature space.In this model,a hypersphere is firstly determined for each sample in the feature space,whose center is this sample and radius d is given ahead.And then the fuzzy membership is determined based on the number of samples locating in the hypersphere.Introducing the novel fuzzy membership model into Adaptive Support Vector Machine （ASVM）,we propose an Adaptive Fuzzy Support Vector Machine algorithm （AFSVM）,Experimental results show that the AFSVM algorithm is valid for improving the anti-noise capacity and the predicting accuracy of the ASVM.