针对传统半监督支持向量机的高斯核函数无法恰当描述流形数据特性,从而导致流形数据分类精度下降的问题,提出一种基于谱聚类的聚类核半监督支持向量机.利用谱聚类方法在特征向量空间中对原始样本数据进行重新表述,使得在新表述中同一聚类中的样本能够更好地积聚在一起,构建聚类核函数,并进而构造聚类核半监督支持向量机,使样本更好地满足半监督学习必须遵循的聚类假设.研究结果表明:聚类核半监督支持向量机对未标记样本的分类精度高且算法性能稳定,对控制参数的设置不敏感,适于解决流形数据的分类问题. Aiming at the problem that the Gaussian kernel function of traditional semi-supervised SVM can not properly describe the manifold data characteristics and lead to the decline of manifold data classification accuracy, a clustering kernel semi-supervised SVM based on spectral clustering is proposed. The class method restates the original sample data in the eigenvector space so that the samples in the same cluster can be better integrated in the new expression to construct the clustering kernel function and construct the clustering kernel semi-supervised support vector Machine to make the sample better meet the clustering assumptions that the semi-supervised learning must follow.The results show that the clustering kernel semi-supervised SVM has high classification accuracy and stable performance for the unlabeled samples and insensitive to the control parameters , Suitable for solving manifold data classification problem.