基于遗传算法的多核支持向量机的参数优化  

Parameters Optimization of Multi-Kernel Support Vector Machine Based on Genetic Algorithm

作　　者：万源[1] 童恒庆[1] 朱映映[2] 

机构地区：[1]武汉理工大学理学院,湖北武汉430070 [2]深圳大学计算机与软件学院,广东深圳518060

出　　处：《武汉大学学报（理学版）》2012年第3期255-259,共5页Journal of Wuhan University:Natural Science Edition

基　　金：国家自然科学基金(60773210);武汉理工大学自主创新基金(20101a037)资助项目

摘　　要：提出了利用遗传算法对多核支持向量机的权系数进行寻优的方法GA-MKL,先选择表现能力最好的参数的单核构成多核,再利用遗传算法来对多核的核系数来寻优.采用该算法在UCI标准数据集上进行了实验,结果表明,该算法为多核SVM的系数选择提供了一种可行的方法.与单核SVM相比,该方法具有更好的分类能力,和其他多核学习算法相比,性能也有一定的提高.In this paper we present a method, GA-MKL, by using the genetic algorithm to optimize the weights of linear multi-kernel support vector machine. We first implement single-kernel experiment to decide the best parameter of each kernel then construct the linear multi-kernel with these indeterminate weights. Then we utilize the genetic al- gorithm to find the best weights which give the best classification performance. The classification experiments on the UCI database are employed with this algorithm. By comparison with the single-kernel algorithm, the experimental results show that this algorithm is superior to the single-kernel, thus it provides a feasible method for finding the best weights for multi-kernel SVM. This algorithm performs better than other MKL algorithm as well.

关 键 词：多核支持向量机 核函数 遗传算法 参数优化 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]