一种比例混合模式特征选择分类器的研究
首发时间:2007-07-24
摘要:本文针对filter模式特征选择方法准确性不高,wrapper模式特征选择方法时间代价很高的缺点,将两者有机的结合一体,设计了一种比例混合模式特征选择分类器。该分类器首先进行filter模式的特征选择,然后根据比例系数挑选其中适应度值较高的个体进行wrapper模式下的特征选择,因此能够取得平衡性较好的特征选择效果。实验选择UCI数据库中的两个数据集为选择对象,链式智能体遗传算法为搜索算法,距离测度为filter模式下的评价准则,RBF神经网络为wrapper模式下的分类器进行了多次实验,统计结果表明,本文设计的特征选择分类器能取得选择速度接近filter模式,识别正确率接近wrapper模式的满意效果。
For information in English, please click here
The Research of One Proportional Hybrid Mode Feature Selection Classifier
Abstract:according to the low accuracy of filter mode feature selection method and high time cost of wrapper mode feature selection method, this paper put them together to propose one proportional hybrid mode feature selection classifier. Firstly, the filter mode feature selection is adopted, then the individuals with higher fitness value are picked out to be put into the classifier with wrapper mode, so the method can obtain the well balanced feature selection result. In the experiment, two datasets in UCI database are chose, chain-like genetic algorithm are chose for searching tool, distance measurement are chose for evaluation rule with filter mode, RBF neural network is chose for feature selection with wrapper mode and classification. The experimental results show that the feature selection classifier can obtain the speed much faster than filter mode, can obtain the satisfied classification error rate as well as wrapper mode.
Keywords: proportion hybrid mode feature selection classifier neural network
论文图表:
引用
No.1422775160118527****
同行评议
共计0人参与
勘误表
一种比例混合模式特征选择分类器的研究
评论
全部评论0/1000