摘要
支撑矢量机(SVM)模型的自动选择是其实际应用的关键.常用的基于穷举搜索的留一法(LOO)很繁杂且效率很低.到目前为止,大多数的算法并不能有效地实现模型自动选择.本文利用实值编码的启发式遗传算法实现基于高斯核函数的SVM模型自动选择.在重点分析了SVM超参数对其性能的影响和两种SVM性能估计的基础上,确定了合适的遗传算法适应度函数.人造数据及实际数据的仿真结果表明了所提方法的可行性和高效性.
Motivated by the facts that automatic model selection for support vector machine （SVM） is an important issue to make it practically useful, and the commonly-used leave-one-out （LOO） method is complex and time consuming, we proposed an effective strategy for automatic model selection for SVM with Gauss kernel by using a heuristic real-coded genetic algorithm （GA）. Based on the extensive analysis of the effects of the hyper-parameters on the generalization performance and two estimates of SVM, the appropriate fitness function for GA operation is determined. Simulations are performed on both artificial data and real data to demonstrate the effectiveness and efficiency of the proposed approach. The significance of the proposed method is its easy implementation and better performances in comparison with the commonly used loo method.
出处
《控制理论与应用》
EI
CAS
CSCD
北大核心
2006年第2期187-192,共6页
Control Theory & Applications
基金
国家自然科学基金资助项目(60133010
60372047)
西安电子科技大学博士点基金资助项目
西安电子科技大学青年工作站项目资助项目
