摘要
针对传统支持向量机(Support Vector Machine,SVM)集成学习(Ensemble Learning,EL)方法不能够解决高维复杂数据且子学习器差异性小集成效果不明显的问题,提出一种基于多种特征选择方法进行Bagging集成的支持向量机学习(Support Vector M achine Based on M ultiple Feature Selection Bagging,M FSB_SVM)方法.该方法首先采用不同的特征选择方法构建子学习器,以增加不同子学习器间的差异性,并直接从训练数据中对样本特征的重要性进行评估,而无需学习算法的反馈.实验表明,本文提出的MFSB_SVM方法既可以有效解决高维数据问题,也可避免传统SVM集成方法效果不明显的缺点,从而进一步提高学习模型的泛化性能.
An SVM bagging ensemble learning algorithm based on feature selection is proposed. The differences of sub-SVM learners are enhanced by different feature selection techniques, and the iteration process of SVM learning is not necessary because the impor- tance of each feature can be directly estimated from the given data. Experimental results on benchmark datasets demonstrate that the proposed approach can not only solve the high dimensional problems, but also avoid the drawback of bagging on SVM learning. In so doing ,the generalization performance can be improved efficiently.
出处
《小型微型计算机系统》
CSCD
北大核心
2014年第11期2533-2537,共5页
Journal of Chinese Computer Systems
基金
国家自然科学基金项目(60975035
61273291
61273294)资助
山西省回国留学人员科研基金项目(2012-008)资助
山西省科技厅基础条件平台项目(2012091003-0104)资助
山西省研究生教育创新项目(20133001)资助
