摘要
支持向量机参数是影响其性能的重要因素,但对支持向量机核参数的选取仍没有形成一套成熟的理论,从而严重影响了其广泛的应用。将克隆选择算法引入差分进化算法,对基本克隆选择算法和差分进化算法中的策略进行改进。将两种改进的算法进行融合,提出了一种基于克隆选择的差分进化算法,并将其应用于SVM核参数的优化中。测试结果表明,该算法不仅可以有效避免差分进化算法易早熟收敛的问题,而且寻优能力得到显著提高;在UCI数据库wine数据中的应用表明,利用克隆选择差分进化算法优化SVM核参数加快了参数搜索的速度,提高了SVM预测精度和泛化能力,具有较高的分类准确率和较好的推广性能。
The parameters of support vector machine （SVM） are important factors affecting its performance. However, the absence of a mature theory about the kernel parameter selection of SVM heavily affects its wide application. This paper introduced clonal selection algorithm into differential evolution algorithm, and improved the strategies of basic clonal selection algorithm and differential evolution algorithm. Through combining the two algorithms mentioned above, a differential evolution algorithm based on clonal selection was proposed and applied to optimize the parameters of SVM kernel. The test results show that the algorithm can not only effectively avoid the premature-convergence problem of differential evolution algorithm, but also significantly improve the optimization ability. UCI wine database application data show that the algorithm can accelerate the parameter search speed, and improve the prediction accuracy and genera- lization ability of SVM. The high accuracy of classification and better generalization per[ormance prove that using clonal selection differential evolution algorithm is a good way to optimize SVM kernel parameter.
出处
《计算机科学》
CSCD
北大核心
2015年第B11期19-21,48,共4页
Computer Science
关键词
克隆选择
差分进化
支持向量机
核参数
Clone selection, Differential evolution, SVM, Kernel parameter
