基于主成分分析的模糊支持向量分类器

Fuzzy support vector machine classifier based on principal component analysis

由于孤立点和原始样本的选取对于支持向量机的分类性能具有较大的影响,所以本文旨在设计一种区别于以往的支持向量机的算法来解决这个问题。首要步骤是通过主成分分析法对原始数据样本进行处理,以达到用最好的方式对原始数据进行表达,从而达到使高维特征空间的维数降低的目的。然后使用类均值法,依据样本在特征空间的投影到特征空间中本类样本均值的距离,来确定其模糊隶属度,为达到使孤立点对最优分类超平面的影响最小,本文通过赋予较小隶属度的方式来实现这一目标。通过进行仿真实验,我们可以发现,这种算法能够比较有效地降低分类误差,而且在一定程度上也能够使支持向量机的鲁棒性得到提高。

A novel support vector machines method is proposed to eliminate the impact of outliers and the original data representation form on the performance of support vector machines. Firstly, in order to decrease the dimensions in high-dimension feature space, principal component analysis is used to deal with the original data. According to the distance to the mean of the same class in the feature space, the membership value of each sample is determined. Then the outliers are given low membership values to depress their influence on the discriminate function. Simulation results show that the presented method reduces the classification error efficiently and improve the robustness of support vector machines.