Journal of Frontiers of Computer Science and Technology ›› 2020, Vol. 14 ›› Issue (7): 1194-1199.DOI: 10.3778/j.issn.1673-9418.1905073

Previous Articles     Next Articles

Granular Support Vector Machine Algorithm Based on Affinity Propagation

CHENG Fengwei, WANG Wenjian   

  1. 1. Department of Computer Engineering, Taiyuan University, Taiyuan 030032, China
    2. School of Computer and Information Technology, Shanxi University, Taiyuan 030006, China
    3. Key Laboratory of Computational Intelligence and Chinese Information Processing of Ministry of Education, Shanxi University, Taiyuan 030006, China
  • Online:2020-07-01 Published:2020-08-12

基于近邻传输的粒度SVM算法

程凤伟王文剑   

  1. 1. 太原学院 计算机工程系,太原 030032
    2. 山西大学 计算机与信息技术学院,太原 030006
    3. 山西大学 计算智能与中文信息处理教育部重点实验室,太原 030006

Abstract:

The granular support vector machine (GSVM) can effectively improve the learning efficiency of support vector machine (SVM) but may lose some generalization ability at same time, because it is sensitive to the initial granulation parameter and the selection of granular centers is rough. This paper proposes a new granular support vector machine model, called affinity propagation based granular support vector machine (APG_SVM). Firstly,a group of high-quality and more representative granular centers are selected to join the training dataset using affinity propagation. And then the training dataset is optimized according to the mixing degree of sample in the granular and the distance between the granular centers and hyperplane. The final training dataset is generated, and the generalization performance of GSVM can be improved by training on the final dataset. The experimental results on UCI standard datasets show that compared with traditional GSVM, the classification efficiency of this algorithm is obviously improved, the accuracy on several datasets is relatively stable, and the classification performance is better.

Key words: granular center, affinity propagation, mixing degree, classification performance

摘要:

传统粒度支持向量机(GSVM)模型可以有效提高支持向量机(SVM)的学习效率,但因其对初始粒划参数比较敏感,粒中心的选取比较粗糙,会损失一定的泛化能力。提出一种基于近邻传输的粒度支持向量机学习算法(APG_SVM)。首先在训练数据上采用近邻传输思想选取一组高质量的更具有代表性的粒中心加入到训练集,再根据粒中样本的混合度及粒中心到超平面的距离对训练集进行优化,生成最终训练集,然后进行训练,这样可使GSVM具有更好的泛化能力。在UCI标准数据集上的实验结果表明,与传统的粒度支持向量机相比,该算法分类效率有明显提高,在几个数据集上的正确率相对稳定,获得了较好的分类性能。

关键词: 粒中心, 近邻传输, 混合度, 分类性能