基于遗传算法的双子支持向量机的模型选择

Genetic algorithm based model selection of TWSVM

双子支持向量机是在支持向量机的基础上提出的一种新的机器学习方法。与传统支持向量机相比,双子支持向量机寻找的是一对不平行的超平面,计算效率是传统支持向量机的4倍。然而,双子支持向量机的参数较多,在应用过程中存在较大局限性。在研究了惩罚参数和核参数对双子支持向量机分类性能影响的基础上,利用遗传算法来选择双子支持向量机的参数,避免了盲目的模型选择。实验结果显示,所提算法能有效选择合适参数,获得的参数能使双子支持向量机具有较好的泛化性能,同时也更加高效。

Twin support vector machine(TWSVM)is a new machine learning method based on support vector machine.Incomparison with traditional support vector machine,TWSVM looks for a pair of non-parallel hyperplane,and its computationalefficiency is increased by3times.However,the parameters of TWSVM are various,and have some limitations in the application process.On the basis of the research on the influence of the penalty parameter and kernel parameter on classification performance of TWSVM,the genetic algorithm is used to select the parameters of TWSVM to avoid blind model selection.The experimental results show that the algorithm can select appropriate parameters effectively,which can make the TWSVM have highgeneralization performance and efficient performance.