An improved rough margin-based v-twin bounded support vector machine
An improved rough margin-based v-twin bounded support vector machine
复制标题
一种改进的基于粗糙边缘的v型孪生有界支持向量机
DOI：
10.1016/j.knosys.2017.05.004
复制
发表时间：
2017
影响因子：
8.8
通讯作者：
Zhou Zhijian
中科院分区：
文献类型：
--
作者：
Wang Huiru;Zhou Zhijian
基于粗糙边缘的ν-twin支持向量机（Roughν-TSVM）在构造分类超平面时，根据误分类样本的位置，对误分类样本给予不同的惩罚，大大提高了测试精度.然而，在求解对偶问题时，它涉及昂贵的矩阵逆运算。在粗糙ν-TSVM的基础上，提出了一种改进的基于粗糙边缘的ν-twin有界支持向量机（I-rough ν-TBSVM）.类似地，所提出的I粗糙ν-TBSVM根据样本的位置给出不同的惩罚。此外，它通过引入正则化项实现了结构风险最小化原则。因此，I粗糙ν-TBSVM比粗糙ν-TSVM具有更高的测试精度。值得一提的是，本文提出的I粗糙ν-TBSVM巧妙地避免了矩阵求逆运算，降低了计算复杂度，节省了更多的运行时间。此外，在非线性情况下，核技巧可以直接应用于I粗糙ν-TBSVM，这对于获得更好的分类性能是必不可少的。该算法具有更好的泛化性能和更高的灵活性。在35个基准数据集上的数值实验验证了算法的有效性。实验结果表明，该算法比同类算法具有更好的性能。
In the rough margin-basedν-twin support vector machine (roughν-TSVM), different penalties are given to the corresponding misclassified samples according their positions when constructing the separating hyperplane, which greatly improved the testing accuracy. However, it involves an expensive matrix inverse operation when solving the dual problem. In this paper, we propose an improved rough margin-basedν-twin bounded support vector machine (I roughν-TBSVM) which is motivated by the roughν-TSVM. Similarly, the proposed I roughν-TBSVM gives different penalties according to the samples' positions. Besides, it implements structural risk minimization principle by introducing a regularization term. So the I roughν-TBSVM yields higher testing accuracy in comparison with roughν-TSVM. It is worthwhile to mention that the proposed I roughν-TBSVM skillfully avoids the matrix inverse operation, which reduces the computational complexity and saves more running time. In addition, the kernel trick can be applied directly to the I roughν-TBSVM for the nonlinear case, which is essential to obtain the better classification performance. It is more flexible and has better generalization performance. Numerical experiments on thirty-five benchmark datasets are performed to investigate the validity of our proposed algorithm. Experimental results indicate that our algorithm gains better performance than the compared algorithms.