一种提高支持向量机针对低维向量分类精度的新方法

本文提出一种支持向量机的全局优化训练算法,形成一种新的分类器,以解决传统的支持向量机在对低维样本点分类时产生的精度下降问题。首先对支持向量机原理,以及以SVM—light为代表的经黄SVM训练算法进行分析,发现支持向量机的训练在本质上都归结为具有不等式约束条件的二次规划问题。本文直接根据支持向量机的最优分类超平面,将其化为无约束条件的求解函数极值问题。然后采用全局优化算法-禁忌搜索算法得到函数的极值点。通过两类高斯样本点分类实验和人脸图像识别的多类分类试验,证明使用支持向量机的全局优化训练算法,在样本点特征向量维数较低的情况下,比使用传统的支持向量机训练算法分类具有更高的分类准确率。

A Novel Method of Improving Accuracy of SVM in Classifying Low-Dimension Vectors

A novel global optimizing training algorithm for Support Vector Machines is proposed in this paper to form a new classifier which improves the accuracy of classic SVM in processing samples with low-dimension vector. The quadratic programming problem with inequality constraint conditions in classic SVM is transformed into solving minimum of object function with unconstrained condition based on the optimal separating hyperplane. The global optimizing algorithm-Tabu searching algorithm is used to get the minimum value. Experiments prove that the training algorithm for SVM in this paper gets the higher accuracy than classic training algorithm in classifying samples with low-dimension vector.