Journal of Frontiers of Computer Science and Technology ›› 2020, Vol. 14 ›› Issue (4): 628-636.DOI: 10.3778/j.issn.1673-9418.1905024

Previous Articles     Next Articles

Grey Wolf Optimizes Mixed Parameter Multi-Classification Twin Support Vector Machine

ZHOU Guangyue, LI Kewen, LIU Wenying, SU Zhaoxin   

  1. College of Computer Science and Technology, China University of Petroleum, Qingdao, Shandong 266580, China
  • Online:2020-04-01 Published:2020-04-10

灰狼优化的混合参数多分类孪生支持向量机

周广悦李克文刘文英苏兆鑫   

  1. 中国石油大学(华东) 计算机科学与技术学院,山东 青岛 266580

Abstract:

Twin support vector machine (TWSVM) is an efficient binary classification algorithm based on support vector machine (SVM). Since most of the problems in reality are multi-classified, it is very important to extend binary classification twin support vector machine to multi-classification twin support vector machine (MTWSVM). At present, the commonly used MTWSVM is generally based on the “one-versus-one” strategy, but each sub-classifier uses the same penalty parameters and core parameters, ignoring the differences between different sub-classifiers, so it cannot play its best role. This paper proposes a multi-classification twin support vector machine based on mixed parameters (MP-MTWSVM). This algorithm selects appropriate parameters for different sub-classifiers, maintaining the diversity of classifiers, then constructing MTWSVM in terms of the “one-versus-one” strategy. TWSVM  faces the problem that its parameters are difficult to be appointed, and MP-MTWSVM algorithm introduces a large number of parameters. This paper optimizes the parameters of MP-MTWSVM by using grey wolf optimizer algorithm (GWO), and further proposes a mixed parameter multi-classification twin support vector machine based on grey wolf optimizer (GWO-MP-MTWSVM). Experiments show that GWO can quickly find the optimal parameters of sub-classifiers and further improve the accuracy of the algorithm.

Key words: multi-classification twin support vector machine (MTWSVM), one-versus-one strategy, mixed parameter, grey wolf optimizer (GWO)

摘要:

孪生支持向量机(TWSVM)是在支持向量机(SVM)的基础上产生的一种高效二分类算法,由于现实中存在的问题大多数是多分类的,将二分类孪生支持向量机扩展到多分类孪生支持向量机(MTWSVM)是非常重要的。目前常用的MTWSVM一般是基于“一对一”策略,但该策略中各子分类器都采用相同的惩罚参数以及核参数,忽略了不同子分类器之间的差异,不能使其发挥最好的作用。通过提出一种基于混合参数的多分类孪生支持向量机(MP-MTWSVM),为不同的子分类器选取合适的参数,保持分类器的多样性,进而根据“一对一”策略构建MTWSVM。TWSVM本就面临着参数难确定的问题,而MP-MTWSVM算法又引入了大量的参数,通过灰狼算法(GWO)对MP-MTWSVM的参数进行寻优,进一步提出了基于灰狼优化的混合参数多分类孪生支持向量机(GWO-MP-MTWSVM)。通过实验表明,GWO可以快速找到各子分类器的最优参数,并进一步提升了算法的准确率。

关键词: 多分类孪生支持向量机(MTWSVM), 一对一策略, 混合参数, 灰狼优化(GWO)