摘要： 提出一种基于数据属性重要性排序的神经网络属性选择方法。该方法只需对部分属性进行训练,即可进行降维。它克服了现有的神经网络降维方法必须对全部属性进行训练的弊端,大大提高了属性选择的效率。该方法先用本文提出的一种简单的可分性判据方法对数据属性进行重要性排序,然后按重要次序用RBF神经网络进行属性选择。仿真实例表明,该方法具有良好的效果。

关键词: 数据降维, 属性重要性排序, 可分性判据, RBF神经网络, 属性选择

Abstract: This paper presents a method of neural networks feature selection based on data attributes importance ranking. Only partial attributions are needed to train a network and the data dimension can be largely reduced by this method .It avoids the deficiency of existing neural networks methods that needs to train all attributions,which improves the efficiency of attributions selection greatly. First, this paper presents a method of data attributes importance ranking using separability criterion. Then, attributes selection is carried out on the basis of attributes importance ranking using the RBF neural network.The efficiency of this approach is demonstrated by simulation examples.

中图分类号: 

• TP183