针对支持向量机对噪声的敏感,以及当两类训练样本数量差别悬殊时,造成分类结果倾向较大类等弱点,通过理论分析,合理地设计隶属度函数,提出了一种新隶属度函数的模糊支持向量机。该方法既可补偿倾向性造成的不利影响,又可增加抗噪声能力,提高预测分类精度。最后通过对含噪声的非均衡数据实验表明,该方法比传统支持向量机和简单去噪模糊支持向量机都有着较高的分类能力。
Since SVM is sensitive to noises or outliers in the training set and the classification of unbalance data is unfair to the rare class,a new fuzzy Support Vector Machine is presented with theoretical analysis given.By properly designing a new fuzzy membership function,the proposed algorithm can compensate the ill-effect of tendency and also can strengthen the ability to detect noises thus improves the accuracy.Simulations on unbalenced data with noise show that,compared with traditional SVM and FSVM,this algorithm has better classification ability.