摘要
为提高Adaboost算法迭代过程中生成基分类器的分类精度以及简化整个集成学习系统的复杂度,文章提出了一种优化Adaboost迭代过程的SVM集成算法。该算法提出了一种在其迭代过程中加入样本选择和特征选择的集成方法。通过均值近邻算法对样本进行选择,并利用相对熵法进行特征选择,最后利用优化得到的特征样本子集对基分类器SVM进行训练,并用加权投票法融合各个SVM基分类器的决策结果进行最终判决。通过对UCI数据集的仿真结果表明,本算法与支持向量机集成算法相比,能够在更少的样本以及特征的基础上,实现较高的识别正确率。
In order to improve the classification accuracy of generating base classifier in Adaboost algorithm iterative process and simplify the complexity of the whole integrated learning system, this paper puts forward an integrated algorithm of SVM to optimize Adaboost iterative process. An integrated approach of adding sample selection and feature selection is proposed during its iteration process. The mean nearest neighbor algorithm is adopted to select the samples, and relative entropy method is used for feature selection, finally the optimized feature sample set is used for training the base classifier SVM, the weighted voting method is finally decided by fusing the decision results of each SVM based classifier. Through the simulation result of UCI data sets, compared with the algorithm integrated with support vector machine algorithm, the algorithm can achieve a high correct recognition rate based on fewer samples as well as features.
出处
《无线互联科技》
2017年第15期106-108,共3页
Wireless Internet Technology
关键词
集成学习
均值近邻
支持向量机
ensemble learning
mean nearest neighbor
support vector machines
