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。 This paper proposes a neural network attribute selection method based on the importance of data attributes. The method only needs to train some attributes to reduce the dimension. It overcomes the disadvantages that the existing methods for dimensionality reduction of neural networks must train all the attributes, and greatly improves the efficiency of attribute selection. The method first uses the simple separability criterion method proposed in this paper to sort the importance of the data attributes, and then uses the RBF neural network to select the attributes in an important order. Simulation examples show that this method has good effect.