【摘要】 提出一种概率神经网络样例选择算法,它包括两个阶段,第一个阶段利用概率神经网络计算样例的不确定性,第二个阶段利用计算出的不确定性选择样例.与压缩近邻规则、编辑近邻规则、约简近邻规则和迭代过滤算法四种代表性的样例选择算法进行了实验比较,实验结果显示在能力保持框架下,该算法的性能优于这四种方法.本文提出的算法具有下列特点:(1)学习速度快;(2)没有分类器的限制;(3)具有好的泛化能力.更多还原

【Abstract】 This paper presents an instance selection algorithm,which is based on probabilistic neural network,and consists of two phases. The uncertainty of instances is calculated with probabilistic neural network in the first phase,the instances are selected according to their uncertainty in the second phase. We conducted experimental comparisons with four representative instance selection algorithms,which are the condensed nearest neighbor( CNN) rule,the edited nearest neighbor( ENN) rule,the reduced nearest neighbor( RNN)rule,and the iterative case filtering( ICF) algorithm. In the framework of competence preservation,the experimental results showthat the proposed algorithm outperforms the four ones. The proposed algorithm has the following characteristics,( 1) Fast learning speed;( 2) There is no restriction on the selected classifiers;( 3) Good generalization ability.更多还原