摘要
将SVM预测精度看作是一个关于模型参数的不连续的多极值函数,基于改进的免疫网络算法,对SVM的模型参数选择问题进行研究,将免疫网络算法与SVM相结合形成一个AIN-SVM算法。分别对分类和回归数据集进行了测试,结果表明该方法能够更快速地在更大的空间内进行有效搜索,与传统的交叉验证方法相比,在搜索速度与稀疏性上具有较大的优势。
Deeming the SVM prediction accuracy as an inconsecutive multi-extreme function correlated to model parameter,the parameter selection of SVM model was studied based on the improved artificial immune net (AINet) algorithm. The AIN-SVM algorithm, in which the AINet algorithm and SVM are integrated, was proposed. A typical classification dataset and a typical regression dataset were tested with this algorithm respectively, and the results show that, the AIN-SVM can effectually search in a bigger space faster, and surpasses traditional cross-validation method a lot in searching speed and sparsity.
出处
《计算机应用与软件》
CSCD
2009年第9期266-268,共3页
Computer Applications and Software
关键词
支持向量机
参数选择
人工免疫网络
Support vector machines (SVM) Parameter selection Artificial immune net