摘要
为有效提升支持向量机泛化性能,提出了基于差分进化算法和负相关学习的选择性支持向量机集成。通过bootstrap技术产生并训练得到多个独立子SVM,基于负相关学习理论构造适应度函数,既提高子SVM的泛化性能,又增大其之间差异度。利用差分进化算法计算各子SVM在加权平均中的最优权重,选择权值大于一定阈值的部分SVM进行加权集成。实验结果表明,该算法是一种有效的集成方法,能进一步提高SVM的泛化性能。
Selective SVM ensemble based on differential evolution and negative correlation learning is presented to improve the generalization ability of SVM. Many SVMs are produced by bootstrap methods, the fitness function is established based on negative correlation learning to improve generalization and high dissimilarity with others. The weighte of SVM is calculated by differential evolution, then those SVMs with weight larger than a given threshold value are ensembled using weights average. Experimental results show that the algorithm is an effect ensemble method and improves the generalization ability of SVM.
出处
《计算机工程与设计》
CSCD
北大核心
2011年第5期1807-1809,1819,共4页
Computer Engineering and Design
基金
陕西省教育厅自然科学研究基金项目(09JK380)
关键词
差分进化算法
适应函数
负相关学习
支持向量机
选择性集成
differentialevolution （DE）
the fitness function
negative correlation learning
support vector machine
selective ensemble
