摘要
为了获得重要的特征集合,提出了一种基于判别式分析算法和神经网络的特征选择方法。通过最小化扩展互熵误差函数来训练神经网络,这一误差函数的使用减小了神经网络传输函数的导数,降低了输出敏感度。该方法首先利用判别式分析算法得到一个有序的特征队列,然后通过正则化神经网络进行特征的选择,特征选择过程是基于单个特征的移除带来验证数据集上分类误差变化这一原理。与其他基于不同原理的四种方法进行了比较,实验结果表明,利用该算法训练的网络能够获得较高分类准确率。
A new approach for feature selection based on discriminant analysis and regulafization neural network was proposed. The neural network was trained by minimizing an augmented cross-entropy error function. The augmented error function forces the neural network to keep low derivatives of the transfer functions of neurons when learning a classification task. Such an approach reduced output sensitivity to the input changes. Firsdy a feature queue in order could he obtained by using discriminant analysis based feature ranking. Feature selection was based on the reaction of the cross-validation data set classification error due to the removal of the individual features. The approach proposed was compared with four other feature selection methods, each of which banks on a different concept. The algorithm proposed outperforms the other methods by achieving higher classification accuracy on all the problems tested.
出处
《计算机应用》
CSCD
北大核心
2006年第2期433-435,共3页
journal of Computer Applications
基金
国家自然科学基金资助项目(60135010)
关键词
特征选择
神经网络
判别式分析
正则化
分类
feature selection
neural network
discriminant analysis
regularization
classification