摘要
为提高支持向量机(SVM)算法的分类精度,本文基于SVM分类算法工作原理,提出一种新的样本权值设置方法,并将SVM最大分类间隔因素引入蚁群算法(ACO)中,实现了优化的混合加权核函数选取.首先,依据最大分类间隔决定SVM分类模型潜在分类能力这一原理,基于样本对最大分类间隔的不同贡献自适应地为其设置权值.然后,将SVM最大分类间隔因素引入ACO搜索算法的参数设置中,对混合加权核函数方案进行确定.本文算法从提高SVM分类模型分类确定性的角度出发,实现了训练样本权值、核函数以及其相应系数的自适应设置、选取.最终,本文方法用于一系列有针对性的笔迹验证实验,实验结果证明用本文方法学习所得SVM分类模型对后续待检测样本具有更高分类精度.
To heighten the classification accuracy of SVM （Support Vector Machine） algorithm,this paper proposes a novel sample weighting algorithm based on the SVM algorithm principle.Meanwhile,an optimal mixed weighting kernel function selection by ACO （Ant colony optimization） is achieved by introducing the maximum classification interval factor into ACO.Firstly,according to the principle that the maximum classification interval will determines the potential classification ability of the learnt SVM classifier model,the instances are adaptively weighted based on their contributions to the maximum classification interval.Then,this paper innovatively introduces the maximum classification interval factor into ACO parameter setting to realize the optimal kernel functions and their corresponding coefficients search.In this paper the choices for kernel functions,their corresponding coefficients and training samples′ weighs are all decided adaptively and optimally on the basis of enhancing the designed SVM classifier model′s classification accuracy and potential.Finally,the proposed method is used in the Chinese handwriting verification and a series of targeted experiments are designed and conducted.The experiment results prove the learnt SVM classifier model possesses a higher classification accuracy for the future testing samples by the proposed method in this paper.
出处
《小型微型计算机系统》
CSCD
北大核心
2018年第2期340-346,共7页
Journal of Chinese Computer Systems
基金
北京首批13所高校高精尖创新中心基金项目(PXM2016_014204_500072)资助
关键词
核函数
SVM
ACO
分类器
kernel function SVM （Support Vector Machine） ACO （Ant colony optimization） classifier
