Kernel discriminant analysis based feature selection
Kernel discriminant analysis based feature selection
复制标题
DOI：
10.1016/j.neucom.2008.02.018
复制
发表时间：
2008-08
期刊：
影响因子：
6
通讯作者：
Tsuneyoshi Ishii;Masamichi Ashihara;S. Abe
中科院分区：
文献类型：
--
作者：
Tsuneyoshi Ishii;Masamichi Ashihara;S. Abe
对于两类问题，我们提出了两个基于核判别分析（KDA）的特征选择标准。第一个是核判别分析的目标函数，称为KDA准则。我们表明，KDA标准是单调的删除功能，这确保了稳定的功能选择。第二个是由KDA分类器获得的识别率，称为基于KDA的识别率，其定义在由KDA获得的一维空间中。即，计算给定类别的数据的条件概率，并且将数据分类到具有最大条件概率的类别中。为了确保稳定的特征选择，我们通过交叉验证来评估基于KDA的识别率。通过计算机实验，我们比较了两个标准的两类问题和交叉验证评估的支持向量机（SVM）的识别率，称为基于SVM的识别率。KDA准则和基于KDA的识别率的选择性能相当，并且优于基于SVM的识别率。
For two-class problems we propose two feature selection criteria based on kernel discriminant analysis (KDA). The first one is the objective function of kernel discriminant analysis called the KDA criterion. We show that the KDA criterion is monotonic for the deletion of features, which ensures stable feature selection. The second one is the recognition rate obtained by a KDA classifier, called the KDA-based recognition rate, which is defined in the one-dimensional space obtained by KDA. Namely, a conditional probability of a datum for a given class is calculated and the datum is classified into the class with the maximum conditional probability. To ensure stable feature selection, we evaluate the KDA-based recognition rate by cross-validation. By computer experiments we compare the two criteria for two-class problems and the recognition rate of the support vector Machine (SVM) evaluated by cross-validation, called the SVM-based recognition rate. The selection performance of the KDA criterion and the KDA-based recognition rate is comparable and is better than that by the SVM-based recognition rate.