一类新样条光滑支持向量机

A new spline smooth support vector machine

对光滑函数的逼近性能进行研究。为了更好地逼近支持向量机模型中的加号函数,提出两个新的三阶样条光滑函数。用新的光滑函数替代支持向量机模型中的加号函数,得到一种新的光滑支持向量机模型。由于新模型具有二阶光滑性,可用Newton-Armijo算法快速求解,数据实验表明新模型具有比较优越的分类性能。

In this paper, approximation performance of smooth function is studied. Two new three-order spline smooth functions are proposed to smooth the plus function in support vector machine （SVM） model in order to improve approximate precision. A new three order spline smooth support vector machine （NTSSVM） is obtained through the replacement of the plus func- tion in SVM with the new smooth functions. NTSSVM is second-order smooth and thus it can be solved by the fast Newton-Armijo algorithm. Numerical results demonstrate that NTSSVM has the better classification capability.