光滑分段孪生支持向量机

Smooth piecewise twin support vector machine

为了解决Sigmoid的积分函数对正号函数的逼近精度低的问题,引入一种具有更强逼近正号函数能力的光滑函数即分段函数,提出了光滑分段孪生支持向量机,并用快速Newton-Armijo算法对其求解。在NDC和UCI数据集上的实验结果表明:光滑分段孪生支持向量机能够有效地处理大规模和高维度数据,且分类精度和分类速度与光滑孪生支持向量机相比得到了改进。

To solve the problem of low approximation precision of integral function of sigmoid function, a piecewise function is introduced, which has stronger ability of smooth function of smooth piecewise function to approximate plus function. Smooth piecewise twin support vector machine （SVM）is proposed. Meanwhile, the fast Newton-Armijo algorithm is used to solved the smooth piecewise twin SVM. Experimental results on NDC and UCI datasets show that smooth piecewise twin SVM can effectively deal with large-scale and high-dimensional data, and classification precision and classification speed of smooth piecewise twin SVM are improved than smooth twin SVM.