一种新的支持向量机混合集成算法

针对基于Boosting和Bagging的集成算法不能有效提高"强学习器"泛化性能的问题,融合Boosting的样本扰动和快速核独立分量分析的特征扰动以生成若干个体支持向量分类器,使生成的训练样本集具有较大的差异性。然后基于模糊核聚类算法根据各个体支持向量机在验证集上的泛化误差选择最优个体进行集成。实验结果表明该算法能进一步提高支持向量机分类器的泛化性能,而且具有较强的稳定性。

A new hybrid SVM Ensemble Algorithm

For the problem that ensemble algorithm based on integrate disturbed samples based on Boosting and features based on Bagging can not effectively improve generalization performance of "strong learning machine",a novel hybrid ensemble algorithm is presented.First,the individual classifiers that have higher difference were trained and generated by boosting and FastKICA.Then,Best individuals are selected by generalization error of svm in the validate set based on KFCM.Finally,the selected members are ensembled.The experiments show that the algorithm have higher generalization performance and strong stability.