基于凸绝对值不等式的半监督最小二乘支持向量机

Semi-supervised Support Vector Machines with Convex Absolute Value Inequalities

针对半监督分类问题,提出了基于凸绝对值不等式的半监督最小二乘支持向量机.传统的半监督支持向量机鲁棒性不强、效率不高,针对这些不足,利用凸绝对值不等式将平面分为两个有重叠的半平面,通过极小化重叠部分以及采用最小二乘支持向量机的思想处理无标签点,提高分类准确率,结果具有一定的鲁棒性.在8个数据集上进行了数值实验,说明了所提出的半监督分类算法的有效性.

For semi-supervised classification problem,We propose semi-supervised least squares support vector machines based on absolute value inequalities.Traditional semi-supervised support vector machines are not robust and efficient.To overcome these shortcomings,by using convex absolute value inequalities,the plane is divided into two overlapping halfspaces.By minimizing the overlap and using least squares support vector machines(LS-SVMs)to process unlabeled points,the training accuracy is improved,and the results are robust.Experimental results on several datasets show the effectiveness of our algorithm.