适合多观测样本的基于LS-SVM的新分类算法

针对多观测样本的二分类问题，提出适合多观测样本的基于LS-SVM的新分类算法。每次分类中，待分类的模式使用多观测样本集进行表示，首先对多观测样本集的标签进行假设，将此假设条件作为LS-SVM中优化问题的约束条件，由此得到分类误差，通过比较两次假设下的分类误差确定多观测样本的类别。该方法无需提前训练获得分类器，而是同时利用已知标签样本和多观测样本集，充分利用同类样本在特征空间中连续分布的特点。最后通过三组实验验证了所提方法的有效性。

Novel LS-SVM based classification algorithm for multi-observation sets

To solve the problem of binary classification based on multi-observation sets, a novel LS-SVM based classification algorithm for multi-observation sets is proposed. In each classification, the object is represented by a multi-observation set, and then it makes an assumption about the class of the multi-observation set. Adding the assumption condition to the constraints of optimization problems in LS-SVM, the class is determined by comparing the different classification errors, which are obtained on different assumptions about the class of the multi-observation set. The method does not require training a classifier before classifications, considerating the labeled samples and multiple observation samples simultaneously and taking advantage of continuity law of similar samples in the feature space. Experiments show that the proposed method is valid and efficient.